Ames_Housing_Prediction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Kaggle House Price Prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting missingpy\r\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/b5/be/998d04d27054b58f0974b5f09f8457778a0a72d4355e0b7ae877b6cfb850/missingpy-0.2.0-py3-none-any.whl (49kB)\r\n",
      "\u001b[K     |████████████████████████████████| 51kB 1.6MB/s \r\n",
      "\u001b[?25hInstalling collected packages: missingpy\r\n",
      "Successfully installed missingpy-0.2.0\r\n"
     ]
    }
   ],
   "source": [
    "# Import useful libraries and change settings\n",
    "import numpy as np # linear algebra\n",
    "import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n",
    "import scipy.stats\n",
    "import re\n",
    "import graphviz\n",
    "import math\n",
    "import IPython\n",
    "import sys\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.linear_model import LinearRegression, Ridge, Lasso, ElasticNet\n",
    "import pdb\n",
    "from random import random\n",
    "from sklearn.ensemble import RandomForestRegressor, ExtraTreesRegressor, GradientBoostingRegressor\n",
    "from sklearn import metrics\n",
    "import sklearn.model_selection as ms\n",
    "from sklearn.experimental import enable_iterative_imputer\n",
    "from sklearn.impute import IterativeImputer\n",
    "from scipy.cluster import hierarchy as hc\n",
    "from sklearn.preprocessing import LabelEncoder, Imputer, StandardScaler\n",
    "from pandas.api.types import is_string_dtype, is_numeric_dtype, is_categorical_dtype\n",
    "from sklearn.ensemble import forest\n",
    "from sklearn.tree import export_graphviz\n",
    "\n",
    "if \"kaggle\" in '\\t'.join(sys.path):\n",
    "    !pip install missingpy\n",
    "else:\n",
    "    import sys\n",
    "    !{sys.executable} -m pip install missingpy\n",
    "    !{sys.executable} -m pip install xgboost\n",
    "\n",
    "import xgboost\n",
    "from missingpy import KNNImputer\n",
    "#from IPython.display import display\n",
    "pd.set_option('display.max_columns', 500)\n",
    "pd.set_option('display.max_rows', 20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Function definitions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Includes trains_cats, apply_cats and proc_df as direct import from FastAI library\n",
    "\n",
    "## For data cleansing\n",
    "\n",
    "def train_cats(df):\n",
    "    for n,c in df.items():\n",
    "        if is_string_dtype(c): \n",
    "            df[n] = c.astype('category').cat.as_ordered()\n",
    "\n",
    "def apply_cats(df, trn):\n",
    "    for n,c in df.items():\n",
    "        if (n in trn.columns) and (trn[n].dtype.name=='category'):\n",
    "            df[n] = c.astype('category').cat.as_ordered()\n",
    "            df[n].cat.set_categories(trn[n].cat.categories, ordered=True, inplace=True)\n",
    "\n",
    "def fix_missing(df, col, name, na_dict):\n",
    "    if is_numeric_dtype(col):\n",
    "        if pd.isnull(col).sum() or (name in na_dict):\n",
    "            df[name+'_na'] = pd.isnull(col)\n",
    "            filler = na_dict[name] if name in na_dict else col.median()\n",
    "            df[name] = col.fillna(filler)\n",
    "            na_dict[name] = filler\n",
    "    return na_dict\n",
    "\n",
    "def numericalize(df, col, name, max_n_cat):\n",
    "    if not is_numeric_dtype(col) and ( max_n_cat is None or len(col.cat.categories)>max_n_cat):\n",
    "        df[name] = pd.Categorical(col).codes+1\n",
    "\n",
    "def proc_df(df, y_fld=None, skip_flds=None, ignore_flds=None, do_scale=False, na_dict=None,\n",
    "            preproc_fn=None, max_n_cat=None, subset=None, mapper=None):\n",
    "    if not ignore_flds: ignore_flds=[]\n",
    "    if not skip_flds: skip_flds=[]\n",
    "    if subset: df = get_sample(df,subset)\n",
    "    else: df = df.copy()\n",
    "    ignored_flds = df.loc[:, ignore_flds]\n",
    "    df.drop(ignore_flds, axis=1, inplace=True)\n",
    "    if preproc_fn: preproc_fn(df)\n",
    "    if y_fld is None: y = None\n",
    "    else:\n",
    "        if not is_numeric_dtype(df[y_fld]): df[y_fld] = pd.Categorical(df[y_fld]).codes\n",
    "        y = df[y_fld].values\n",
    "        skip_flds += [y_fld]\n",
    "    df.drop(skip_flds, axis=1, inplace=True)\n",
    "    if na_dict is None: na_dict = {}\n",
    "    else: na_dict = na_dict.copy()\n",
    "    na_dict_initial = na_dict.copy()\n",
    "    for n,c in df.items(): na_dict = fix_missing(df, c, n, na_dict)\n",
    "    if len(na_dict_initial.keys()) > 0:\n",
    "        df.drop([a + '_na' for a in list(set(na_dict.keys()) - set(na_dict_initial.keys()))], axis=1, inplace=True)\n",
    "    if do_scale: mapper = scale_vars(df, mapper)\n",
    "    for n,c in df.items(): numericalize(df, c, n, max_n_cat)\n",
    "    df = pd.get_dummies(df, dummy_na=True)\n",
    "    df = pd.concat([ignored_flds, df], axis=1)\n",
    "    res = [df, y, na_dict]\n",
    "    if do_scale: res = res + [mapper]\n",
    "    return res\n",
    "\n",
    "# For model interpretation\n",
    "\n",
    "def rf_feat_importance(m, df):\n",
    "    return pd.DataFrame({'cols':df.columns, 'imp':m.feature_importances_}\n",
    "                       ).sort_values('imp', ascending=False)\n",
    "\n",
    "def rmse(x,y): return math.sqrt(((x-y)**2).mean())\n",
    "\n",
    "def se(x,y): return (x-y)**2\n",
    "\n",
    "def mse(x,y): return ((x-y)**2).mean()\n",
    "\n",
    "def print_score(m):\n",
    "    res = [rmse(m.predict(X_train), y_train), rmse(m.predict(X_valid), y_valid), m.score(X_train, y_train), m.score(X_valid, y_valid)]\n",
    "    if hasattr(m, 'oob_score_'): res.append(m.oob_score_)\n",
    "    print(res)\n",
    "\n",
    "# For visualisation\n",
    "\n",
    "def draw_tree(t, df, size=10, ratio=0.6, precision=0):\n",
    "    s=export_graphviz(t, out_file=None, feature_names=df.columns, filled=True,\n",
    "                      special_characters=True, rotate=True, precision=precision)\n",
    "    IPython.display.display(graphviz.Source(re.sub('Tree {',\n",
    "       f'Tree {{ size={size}; ratio={ratio}', s)))\n",
    "    \n",
    "def plot_fi(fi): return fi.plot('cols', 'imp', 'barh', figsize=(12,7), legend=False)\n",
    "\n",
    "def pvalue_sample_differences(dset1, dset2, variable_name):\n",
    "    if dset1[variable_name].dtype != dset2[variable_name].dtype:\n",
    "        raise Exception(\"Choose variables of the same type!\")\n",
    "    elif dset1[variable_name].dtype == \"int64\" or dset1[variable_name].dtype == \"float64\":\n",
    "        return scipy.stats.ks_2samp(dset1[variable_name], dset2[variable_name])[1]\n",
    "    elif dset1[variable_name].dtype == \"object\":\n",
    "        cross_tabl = pd.concat([dset1[variable_name].value_counts(sort = True), dset2[variable_name].value_counts(sort = True)], axis=1, sort=True)\n",
    "        return scipy.stats.chi2_contingency(cross_tab1)[1]\n",
    "    else:\n",
    "        raise Exception(\"Use numerical or categorical data.\")\n",
    "        \n",
    "def split_vals(a,n): return a[:n].copy(), a[n:].copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def numericalize2(df, name):\n",
    "    if not is_numeric_dtype(df[name]):\n",
    "        df[name] = pd.Categorical(df[name]).codes+1\n",
    "\n",
    "def fix_missing2(df, split_point, imputation_methodology):\n",
    "    na_dict = {}\n",
    "    \n",
    "    df_temp = df.iloc[:split_point, :]\n",
    "    \n",
    "    for name, _ in df.items():\n",
    "        if is_numeric_dtype(df[name]) and pd.isnull(df[name]).sum():\n",
    "            df[name+'_NA'] = pd.isnull(df[name]).astype(\"int64\")\n",
    "            if imputation_methodology == \"median\":\n",
    "                filler = df_temp[name].median()\n",
    "            elif imputation_methodology == \"mean\":\n",
    "                filler = df_temp[name].mean()\n",
    "            elif imputation_methodology == \"zero\":\n",
    "                filler = 0\n",
    "            else:   \n",
    "                raise Exception(\"Invalid imputation methodology\")\n",
    "            df[name] = df[name].fillna(filler)\n",
    "            na_dict[name] = filler\n",
    "    return na_dict\n",
    "\n",
    "def fix_missing_alg(df, split_point, imputation_methodology):\n",
    "    na_dict = {}                \n",
    "    df_temp = df.iloc[:split_point, :]\n",
    "    \n",
    "    if imputation_methodology == \"5NN\":\n",
    "        imp = KNNImputer(n_neighbors=5, weights=\"uniform\", col_max_missing = 1, row_max_missing = 1)\n",
    "    elif imputation_methodology == \"KNN\":\n",
    "        imp = KNNImputer(n_neighbors=round(math.sqrt(df.shape[0])), weights=\"uniform\", col_max_missing = 1, row_max_missing = 1)\n",
    "    elif imputation_methodology == \"IterativeImputer\":\n",
    "        imp = IterativeImputer(max_iter=1000, random_state=0)\n",
    "    else:   \n",
    "        raise Exception(\"Invalid imputation methodology\")\n",
    "        \n",
    "    df_combined_temp = pd.get_dummies(df, dummy_na=True) \n",
    "    for variable_name in df_combined_temp.filter(regex = \"nan$\",axis=1).columns:\n",
    "        variable_name_suffix = variable_name[0:variable_name.find('_')]\n",
    "        regex_filter = variable_name_suffix + \"$\"\n",
    "        for suffix_varable in df_combined_temp.filter(regex = regex_filter,axis=1).columns:\n",
    "            df_combined_temp[df_combined_temp[variable_name] == 1, suffix_varable] = np.nan\n",
    "        df_combined_temp.drop([variable_name], axis=1, inplace=True)\n",
    "    \n",
    "    column_names = list(df_combined_temp.columns)\n",
    "    df_train_temp_filled = imp.fit_transform(df_combined_temp.iloc[:split_point, :].copy())\n",
    "    \n",
    "    if df_combined_temp.shape[0] != split_point:\n",
    "        df_test_temp_filled = imp.transform(df_combined_temp.iloc[split_point:, :].copy())\n",
    "        df_combined_filled = pd.concat([pd.DataFrame(df_train_temp_filled, columns = column_names), pd.DataFrame(df_test_temp_filled, columns = column_names)])\n",
    "        df_combined_filled.index = df_combined_temp.index\n",
    "    else:\n",
    "        df_combined_filled = pd.DataFrame(df_train_temp_filled, columns = column_names, index = df.index)\n",
    "        \n",
    "    for name in [\"PropertyType1\", \"ElectricalType\", 'VeneerType', 'VeneerArea', \"LotFrontage\", \"Utilities\", \"KitchenQual\", \"Functional\", \"ZoningClass\", \"Exterior1\", \"Exterior2\", \"SaleType\", \"GarageAge\", \"GarageYrBlt\", \"GarageArea\", \"GarageCarsCount\"]:\n",
    "        if pd.isnull(df[name]).sum():\n",
    "            df[name+'_NA'] = pd.isnull(df[name]).astype(\"int64\")\n",
    "            na_dict[name] = imputation_methodology\n",
    "            if name == \"LotFrontage\":\n",
    "                df[name] = df_combined_filled[name]\n",
    "            elif name in [\"PropertyType1\", \"ElectricalType\", \"VeneerType\", \"Utilities\", \"KitchenQual\", \"Functional\", \"ZoningClass\", \"Exterior1\", \"Exterior2\", \"SaleType\"]:\n",
    "                regex_item = \"^\" + name\n",
    "                variable_list = list(df_combined_filled.filter(regex=regex_item,axis=1).columns)\n",
    "                df[name] = df_combined_filled[variable_list].idxmax(axis=1).str.replace(name + \"_\", \"\")         \n",
    "            elif name == \"VeneerArea\":\n",
    "                df_combined_filled.loc[df[\"VeneerType\"] == \"None\", \"VeneerArea\"] = 0\n",
    "                df[name] = df_combined_filled[name]\n",
    "            elif name in [\"GarageAge\", \"GarageYrBlt\", \"GarageArea\", \"GarageCarsCount\"]:\n",
    "                df_combined_filled.loc[df[\"GarageType\"] == \"None\", name] = df[name].median()\n",
    "                df[name] = df_combined_filled[name]\n",
    "    return na_dict\n",
    "\n",
    "def proc_df2(vars_to_cat, df1, df2 = None, y_fld = \"SalePrice\", n_validation = None, percentage_validation = None, add_variables = False, \n",
    "    imputation_methodology = \"median\", impute_using = \"train\", drop_outliers = False, log_size_variables = False, square_variables = False, drop_variables = False):\n",
    "    \"\"\" proc_df2 takes the Ames data set, performs the requested transformations, and \n",
    "    changes the df into an entirely numeric dataframe.\n",
    "    Parameters:\n",
    "    -----------\n",
    "    vars_to_cat: Options are \"none\", \"all\", and \"ordinal.\" All converts all string variables to categories,\n",
    "    and \"ordinal\" converts variables that on inspect seemed to be ordinal to category variables.\n",
    "    df1: The training data frame that you wish to process.\n",
    "    df2: The test data that you wish to process.  Note that information used in imputing is entirely generated\n",
    "    from the training data set (not the validation or test set) unless imputing_using = \"all\". This option must be None to include a validation set.\n",
    "    y_fld: Options are \"SalePrice\" and \"PricePerArea.\"  If \"PricePerArea\" is used, this variable is generated\n",
    "    from the data set and SalePrice is dropped.\n",
    "    n_validation: This must be None if percentage_validation is not None or if df2 is not None. If this\n",
    "    option equals an integer N, df1 will be split into a train and validation data set, with the number of observations\n",
    "    in the validation data set being equal to N.\n",
    "    percentage_validation: This must be None if n_validation is not None or if df2 is not None. If this\n",
    "    option equals a fraction P, df1 will be split into a train and validation data set, with the proportion\n",
    "    of observations in the validation set being equal to P.\n",
    "    add_variables:  If True, various variables are generated. \n",
    "    imputation_methodology: Options are \"KNN\", \"IterativeImputer\" (MICE), \"median\", \"mean\", and \"zero\".\n",
    "    impute_using: Options are \"train\" and \"all.\"  \"Train\" implies that only data from df1 is used\n",
    "    to impute, while \"all\" uses information from df1 and df2 for the imputation process.\n",
    "    drop_outliers: Options are False, \"5SD\", \"10SD\", and \"manual.\"\n",
    "    log_size_variables: If True, various variables related to the size of the property\n",
    "    that are non-zero for all observations are logged.\n",
    "    square_variables: If True, the square of various variables that were found to be\n",
    "    important in the tree model are added to the data frame.\n",
    "    drop_variables: Options are False, \"minimal\", and \"comprehensive.\"  \"Minimal\" only\n",
    "    drops variables whose information is entirely captured by the variables that are added,\n",
    "    while comprehensive drops a larger number of variables, including some that likely include\n",
    "    additional information.  This option must be False if add_variables if False.\n",
    "    \n",
    "    Returns:\n",
    "    --------\n",
    "    IF DF2 IS NONE, N_VALIDATION IS NONE, AND PERCENTAGE_VALIDATION IS NONE:\n",
    "    [x, y, nas]:\n",
    "        x: x is the transformed version of the entire data set -- all elements will be numeric.\n",
    "        y: y is the response variable\n",
    "        nas: returns a dictionary of which nas it created, and the associated median.\n",
    "        \n",
    "    IF DF2 IS NOT NONE, N_VALIDATION IS NOT NONE, OR PERCENTAGE_VALIDATION IS NOT NONE:\n",
    "    [x_train, y_train, x_valortest, y_valortest, nas]:\n",
    "        x_train: x is the transformed version of the training portion of the data set.\n",
    "        y_train: y is the response variable for the training portion of the data set.\n",
    "        x_valortest: x is the transformed version of the validation/test portion of the data set.\n",
    "        y_valortest: y is the response variable for the validation/test portion of the data set.  \n",
    "        **IT LIKELY DOESN'T EXIST FOR TEST DATA SETS.**        \n",
    "        nas: returns a dictionary of which nas it created, and the associated median.\n",
    "    \"\"\"\n",
    "\n",
    "    # Checking for incompatibilities among types.\n",
    "    if df2 is not None and (n_validation is not None or percentage_validation is not None):\n",
    "        raise Exception(\"Cannot transform two datasets and generate validation data set!\")\n",
    "    if df1.shape[0] == 0 or (df2 is not None and df2.shape[0] == 0):\n",
    "        raise Exception(\"Cannot use empty dataframes!\")\n",
    "    if n_validation is not None and n_validation >= df1.shape[0]:\n",
    "        raise Exception(\"Cannot have validation dataset be larger or equal in size to total data set!\")\n",
    "    if drop_variables is not False and add_variables is False:\n",
    "        raise Exception(\"Don't drop variables (yet) unless you are keeping the information somewhere else!\")\n",
    "    if vars_to_cat == \"ordinal\" and drop_variables == \"comprehensive\":\n",
    "        raise Exception(\"If you are converting variables to category, you are estimating a tree model and don't need to drop a ton of variables!\")\n",
    "\n",
    "    if drop_outliers == \"manual\" and vars_to_cat == \"none\":\n",
    "        print(\"Manually Dropping Size Outliers...\")\n",
    "        df1_copy = df1[np.logical_or(df1[\"AboveGradeAreaSF\"] < 5000, pd.isna(df1[\"AboveGradeAreaSF\"]))].copy()\n",
    "    elif drop_outliers == \"5SD\" and vars_to_cat == \"none\":\n",
    "        print(\"Dropping 5 SD Outliers...\")\n",
    "        df1_copy = df1[(np.abs(scipy.stats.zscore(df1[[\"1stFlrSF\", \"LotArea\"]].select_dtypes(include=['float64', \"int64\"]))) < 5).all(axis=1)].copy()\n",
    "    elif drop_outliers == \"10SD\"  and vars_to_cat == \"none\":\n",
    "        print(\"Dropping 10 SD Outliers...\")\n",
    "        df1_copy = df1[(np.abs(scipy.stats.zscore(df1[[\"1stFlrSF\", \"LotArea\"]].select_dtypes(include=['float64', \"int64\"]))) < 10).all(axis=1)].copy()\n",
    "    else:\n",
    "        df1_copy = df1.copy()\n",
    "    \n",
    "    if df2 is not None:\n",
    "        df2_copy = df2\n",
    "        split_point = df1_copy.shape[0]\n",
    "        df_combined = pd.concat([df1_copy, df2_copy])\n",
    "    elif n_validation is not None:\n",
    "        split_point = df1_copy.shape[0] - n_validation\n",
    "        df_combined = df1_copy\n",
    "    elif percentage_validation is not None:\n",
    "        split_point = round(df1_copy.shape[0] * (1 - percentage_validation))\n",
    "        df_combined = df1_copy\n",
    "    else:\n",
    "        split_point = df1_copy.shape[0]\n",
    "        df_combined = df1_copy\n",
    "        \n",
    "    if impute_using == \"all\":\n",
    "        imputation_split = df_combined.shape[0]\n",
    "    else:\n",
    "        imputation_split = split_point\n",
    "        \n",
    "    for variable_name in [\"FinishedBasementSF1\", \"FinishedBasementSF2\", \"UnfinishedBasementSF\", \"TotalBsmtSF\", \"BsmtFullBathCount\", \"BsmtHalfBathCount\"]:    \n",
    "        df_combined.loc[np.logical_and(df_combined[\"BsmtQual\"] == \"None\", df_combined[\"BsmtCond\"] == \"None\"), variable_name] = 0\n",
    "\n",
    "    if imputation_methodology == \"KNN\" or imputation_methodology == \"IterativeImputer\":\n",
    "        na_dict = fix_missing_alg(df_combined, imputation_split, imputation_methodology)\n",
    "\n",
    "    if add_variables is True:\n",
    "        df_combined[\"HouseAge\"] = df_combined.YrSold - df_combined.YearBuilt\n",
    "        df_combined[\"RemodAge\"] = df_combined.YrSold - df_combined.YearLastRemodeled\n",
    "        df_combined[\"GarageAge\"] = df_combined.YrSold - df_combined.GarageYrBlt\n",
    "    \n",
    "        df_combined[\"Season\"] = \"\"\n",
    "        df_combined.loc[np.logical_or(np.logical_or(df_combined[\"MoSold\"] == 3, df_combined[\"MoSold\"] == 4), df_combined[\"MoSold\"] == 5), \"Season\"] = \"Spring\"\n",
    "        df_combined.loc[np.logical_or(np.logical_or(df_combined[\"MoSold\"] == 6, df_combined[\"MoSold\"] == 7), df_combined[\"MoSold\"] == 8), \"Season\"] = \"Summer\"\n",
    "        df_combined.loc[np.logical_or(np.logical_or(df_combined[\"MoSold\"] == 9, df_combined[\"MoSold\"] == 10), df_combined[\"MoSold\"] == 11), \"Season\"] = \"Fall\"\n",
    "        df_combined.loc[np.logical_or(np.logical_or(df_combined[\"MoSold\"] == 12, df_combined[\"MoSold\"] == 1), df_combined[\"MoSold\"] == 2), \"Season\"] = \"Winter\"\n",
    "        \n",
    "        df_combined[\"Months_from_Start\"] = (df_combined[\"MoSold\"] - 1) + 12*(df_combined[\"YrSold\"] - 2006)\n",
    "        \n",
    "        df_combined[\"Residential\"] = df_combined[\"ZoningClass\"].isin([\"FloatingVillage\", \"ResHighDens\", \"ResLowDens\", \"ResLowDensPark\", \"ResMedDens\"]).astype(int)             \n",
    "        df_combined[\"MixedUse\"] = df_combined[\"ZoningClass\"].isin([\"ResHighDens\", \"FloatingVillage\", \"Commercial\", \"Industrial\"]).astype(int)     \n",
    "        df_combined[\"LowDensity\"] = df_combined[\"ZoningClass\"].isin([\"Agriculture\", \"FloatingVillage\", \"ResLowDens\", \"ResLowDensPark\"]).astype(int) \n",
    "\n",
    "        df_combined[\"Depression_Price_Flags\"] = 0\n",
    "        df_combined.loc[np.logical_or(np.logical_or(np.logical_or(df_combined[\"SaleFlags\"] == \"Abnormal\", df_combined[\"SaleFlags\"] == \"Family\"), df_combined[\"SaleFlags\"] == \"Adjoining\"), df_combined[\"SaleType\"] == \"COD\"), \"Depression_Price_Flags\"] = 1\n",
    "\n",
    "        df_combined[\"Warranty\"] = df_combined[\"SaleType\"].isin([\"WD\", \"CWD\", \"VWD\"]).astype(int) \n",
    "        df_combined[\"New\"] = df_combined[\"SaleType\"].isin([\"New\"]).astype(int) \n",
    "        df_combined[\"Reg_Contract\"] = df_combined[\"SaleType\"].isin([\"Con\"]).astype(int) \n",
    "        df_combined[\"Irr_Contract\"] = df_combined[\"SaleType\"].isin([\"ConLw\", \"ConLI\", \"ConLD\"]).astype(int) \n",
    "        df_combined[\"SaleType_NA\"] = df_combined[\"SaleType\"].isna().astype(\"int64\")\n",
    "\n",
    "        df_combined[\"Pre_War_Original\"] = 0\n",
    "        df_combined.loc[np.logical_and(df_combined[\"YearBuilt\"] < 1946, df_combined[\"YearLastRemodeled\"] < 1946), \"Pre_War_Original\"] = 1\n",
    "        \n",
    "        df_combined[\"Pre_War_Remodelled\"] = 0\n",
    "        df_combined.loc[np.logical_and(df_combined[\"YearBuilt\"] < 1946, df_combined[\"YearLastRemodeled\"] >= 1946), \"Pre_War_Remodelled\"] = 1\n",
    "\n",
    "        df_combined[\"Missing_Utilities\"] = df_combined[\"Utilities\"].isin([\"NoSewer\", \"NoSewerWater\", \"OnlyElectric\"]).astype(int)\n",
    "        df_combined[\"NA_Utilities\"] = df_combined[\"Utilities\"].isna().astype(\"int64\")\n",
    "        \n",
    "        urban_amenities = [\"AdjacentAmenity\", \"CloseAmenit\"]\n",
    "        rail_amenities = [\"CloseNSRail\", \"CloseEWRail\", \"AdjacentNSRail\", \"AdjacentEWRail\"]\n",
    "        street_amenities = [\"ArteryStreet\", \"FeederStreet\"]\n",
    "        \n",
    "        df_combined[\"Number_of_Urban_Amenities\"] = df_combined.isin(urban_amenities).sum(1) \n",
    "        df_combined[\"Number_of_Rail_Amenities\"] = df_combined.isin(rail_amenities).sum(1) \n",
    "        df_combined[\"Number_of_Street_Amenities\"] = df_combined.isin(street_amenities).sum(1) \n",
    "        \n",
    "        df_combined[\"Minor_Defects\"] = df_combined[\"Functional\"].isin([\"MinorDeduction1\", \"MinorDeduction2\"]).astype(int) \n",
    "        df_combined[\"Moderate_Defects\"] = df_combined[\"Functional\"].isin([\"ModerateDeduction\"]).astype(int) \n",
    "        df_combined[\"Major_Defects\"] = df_combined[\"Functional\"].isin([\"MajorDeductions1\", \"MajorDeduction2\"]).astype(int) \n",
    "        df_combined[\"Worst_Defects\"] = df_combined[\"Functional\"].isin([\"SevereDamage\", \"Salvage\"]).astype(int) \n",
    "        df_combined[\"Functional_Missing\"] = df_combined[\"Functional\"].isna().astype(\"int64\")\n",
    "        \n",
    "        df_combined[\"TotalHalfBath\"] = df_combined[\"HalfBathCount\"] + df_combined[\"BsmtHalfBathCount\"]\n",
    "        df_combined[\"TotalFullBath\"] = df_combined[\"FullBathCount\"] + df_combined[\"BsmtFullBathCount\"]\n",
    "        df_combined[\"TotalBath\"] = (0.5*df_combined[\"TotalHalfBath\"]) + df_combined[\"TotalFullBath\"]\n",
    "        \n",
    "        df_combined[\"PorchArea\"] = df_combined[\"WoodDeckSF\"] + df_combined[\"OpenPorchSF\"] + df_combined[\"EnclosedPorchSF\"] + df_combined[\"3SsnPorchSF\"] + df_combined[\"ScreenPorchSF\"]\n",
    "        df_combined['MainPorchType'] = df_combined[['WoodDeckSF','OpenPorchSF', 'EnclosedPorchSF', '3SsnPorchSF', 'ScreenPorchSF']].idxmax(axis=1)\n",
    "\n",
    "        df_combined[\"Gas_Heater\"] = df_combined[\"HeatingType\"].isin([\"GasAir\", \"GasWater\"]).astype(int) \n",
    "        df_combined[\"Steam_Heater\"] = df_combined[\"HeatingType\"].isin([\"OtherWater\"]).astype(int) \n",
    "        df_combined[\"Other_Heater\"] =  df_combined[\"HeatingType\"].isin([\"Floor\", \"Grav\", \"Other_Heater\"]).astype(int) \n",
    "        \n",
    "        df_combined[\"FenceType\"] = \"None\"\n",
    "        df_combined.loc[np.logical_or(df_combined[\"FenceQual\"] == \"GoodPrivacy\", df_combined[\"FenceQual\"] == \"MinimumPrivacy\"), \"FenceType\"] = \"Privacy\"\n",
    "        df_combined.loc[np.logical_or(df_combined[\"FenceQual\"] == \"GoodWood\", df_combined[\"FenceQual\"] == \"MinimumWoodWire\"), \"FenceType\"] = \"Wood\"\n",
    "\n",
    "        df_combined[\"FenceQuality\"] = \"None\"\n",
    "        df_combined.loc[np.logical_or(df_combined[\"FenceQual\"] == \"GoodPrivacy\", df_combined[\"FenceQual\"] == \"GoodWood\"), \"FenceQuality\"] = \"Good\"\n",
    "        df_combined.loc[np.logical_or(df_combined[\"FenceQual\"] == \"MinimumWoodWire\", df_combined[\"FenceQual\"] == \"MinimumPrivacy\"), \"FenceQuality\"] = \"Fair\"\n",
    "        \n",
    "        df_combined[\"All_Paved\"] = 1\n",
    "        df_combined.loc[df_combined[\"StreetMaterial\"] == \"Gravel\", \"All_Paved\"] = 0\n",
    "        df_combined.loc[~(df_combined[\"AlleyMaterial\"] == \"Paved\"), \"All_Paved\"] = 0\n",
    "        df_combined.loc[~(df_combined[\"PavedDriveType\"] == \"Paved\"), \"All_Paved\"] = 0\n",
    "        \n",
    "        df_combined['Excellent_count'] = df_combined.isin([\"Excellent\"]).sum(1) \n",
    "        df_combined['Good_count'] = df_combined.isin([\"Good\"]).sum(1) \n",
    "        df_combined['Average_count'] = df_combined.isin([\"Average\"]).sum(1) \n",
    "        df_combined['Fair_count'] = df_combined.isin([\"Fair\"]).sum(1) \n",
    "        df_combined['Poor_count'] = df_combined.isin([\"Poor\"]).sum(1) \n",
    "        df_combined['Total_Rated'] = df_combined['Excellent_count'] + df_combined['Good_count'] + df_combined['Average_count'] + df_combined['Fair_count'] + df_combined['Poor_count']\n",
    "        \n",
    "        df_combined['Average_Score'] = (9*df_combined['Excellent_count'] + 7*df_combined['Good_count'] + 5*df_combined['Average_count'] + 3*df_combined['Fair_count'] + 2*df_combined['Poor_count']) / df_combined['Total_Rated']\n",
    "        \n",
    "        df_combined[\"Excess_Street_Access\"] = df_combined[\"LotConfig\"].isin([\"Corner\", \"Frontage2Sides\", \"Frontage3Sides\"]).astype(int)  \n",
    "        \n",
    "        df_combined[\"Regular_Slope\"] = 1\n",
    "        df_combined.loc[df_combined[\"LandContour\"] != \"Level\", \"Regular_Slope\"] = 0\n",
    "        df_combined.loc[df_combined[\"LandSlope\"] != \"Gentle\", \"Regular_Slope\"] = 0\n",
    "        \n",
    "        df_combined[\"Irregular_Shape\"] = df_combined[\"LotShape\"].isin([\"IR1\", \"IR2\", \"IR3\"]).astype(int) \n",
    "        \n",
    "        df_combined[\"Neigh1\"] = df_combined[\"Neighborhood\"].isin([\"Northridge\", \"NorthridgeHeights\", \"StoneBrook\"]).astype(int) \n",
    "        df_combined[\"Neigh2\"] = df_combined[\"Neighborhood\"].isin([\"Timberland\", \"Veenker\", \"Somerset\", \"ClearCreek\"]).astype(int) \n",
    "        df_combined[\"Neigh3\"] = df_combined[\"Neighborhood\"].isin([\"Crawford\", \"BloomingtonHeights\", \"Gilbert\", \"NWAmes\"]).astype(int) \n",
    "        df_combined[\"Neigh4\"] = df_combined[\"Neighborhood\"].isin([\"SawyerWest\", \"Mitchell\", \"NorthAmes\", \"NorthparkVilla\"]).astype(int) \n",
    "        df_combined[\"Neigh5\"] = df_combined[\"Neighborhood\"].isin([\"SWISU\", \"Bluestem\", \"Sawyer\"]).astype(int) \n",
    "        df_combined[\"Neigh6\"] = df_combined[\"Neighborhood\"].isin([\"OldTown\", \"Edwards\", \"Brookside\", \"Briardale\"]).astype(int) \n",
    "        df_combined[\"Neigh7\"] = df_combined[\"Neighborhood\"].isin([\"MeadowVillage\", \"IDOTRR\"]).astype(int) \n",
    "        \n",
    "        df_combined[\"Total_Rooms\"] = df_combined[\"AboveGradeRooms\"] + df_combined[\"FinishedBasementType1\"].isin([\"GoodLivingQuarters\", \"AverageLivingQuarters\", \"BelowAverageLivingQuarters\", \"RecRoom\", \"LowQuality\"]).astype(int) + df_combined[\"FinishedBasementType2\"].isin([\"GoodLivingQuarters\", \"AverageLivingQuarters\", \"BelowAverageLivingQuarters\", \"RecRoom\", \"LowQuality\"]).astype(int)\n",
    "\n",
    "        df_combined[\"Total_House_Sqft\"] = df_combined[\"TotalBsmtSF\"] + df_combined[\"AboveGradeAreaSF\"]\n",
    "        df_combined[\"House_Proportion\"] = df_combined[\"1stFlrSF\"]/ df_combined[\"LotArea\"]\n",
    "        df_combined[\"House_Proportion_Alt\"] = df_combined[\"Total_House_Sqft\"]/ df_combined[\"LotArea\"]\n",
    "        df_combined[\"Fin_Bsmt_Sqft\"] = df_combined[\"FinishedBasementSF1\"] + df_combined[\"FinishedBasementSF2\"]\n",
    "\n",
    "        df_combined[\"Single_Exterior\"] = (np.asarray(df_combined.Exterior1) == np.asarray(df_combined.Exterior2)).astype(int) \n",
    "        \n",
    "        df_combined[\"Duplex\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"Duplex\", df_combined[\"PropertyType2\"] == \"Duplex\"), \"Duplex\"] = 1\n",
    "        df_combined[\"PUD\"] = df_combined[\"PropertyType1\"].isin([\"1StoryPlannedDevelopment\", \"1.5StoryPlannedDevelopment\", \"2StoryPlannedDevelopment\", \"MultilevelPlannedDevelopment\"]).astype(int) \n",
    "        \n",
    "        df_combined[\"1Story\"] = 0\n",
    "        df_combined.loc[np.logical_or(np.logical_or(np.logical_or(np.logical_or(df_combined[\"PropertyType1\"] == \"1StoryPostWar\", df_combined[\"PropertyType1\"] == \"1StoryPreWar\"), df_combined[\"PropertyType1\"] == \"1StoryAttic\"), df_combined[\"PropertyType1\"] == \"1StoryPlannedDevelopment\"), df_combined[\"HouseStyle\"] == \"1Story\"), \"1Story\"] = 1\n",
    "        \n",
    "        df_combined[\"1.5UnfStory\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"1.5StoryUnfinished\", df_combined[\"HouseStyle\"] == \"1.5Finished\"), \"1.5UnfStory\"] = 1\n",
    "        df_combined[\"1.5FinStory\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"1.5StoryFinished\", df_combined[\"HouseStyle\"] == \"1.5Finished\"), \"1.5FinStory\"] = 1\n",
    "        \n",
    "        df_combined[\"2Story\"] = 0\n",
    "        df_combined.loc[np.logical_or(np.logical_or(np.logical_or(df_combined[\"PropertyType1\"] == \"2StoryPreWar\", df_combined[\"PropertyType1\"] == \"2StoryPostWar\"), df_combined[\"PropertyType1\"] == \"2StoryPlannedDevelopment\"), df_combined[\"HouseStyle\"] == \"2Story\"), \"2Story\"] = 1\n",
    "        \n",
    "        df_combined[\"2.5Story\"] = 0\n",
    "        df_combined.loc[np.logical_or(np.logical_or(df_combined[\"PropertyType1\"] == \"2.5Story\", df_combined[\"HouseStyle\"] == \"2.5Finished\"), df_combined[\"HouseStyle\"] == \"2.5Unfinished\"), \"2.5Story\"] = 1\n",
    "        \n",
    "        df_combined[\"SplitFoyer\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"SplitFoyer\", df_combined[\"HouseStyle\"] == \"SplitFoyer\"), \"SplitFoyer\"] = 1\n",
    "        \n",
    "        df_combined[\"SplitLevel\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"SplitNoFoyer\", df_combined[\"HouseStyle\"] == \"SplitLevel\"), \"SplitLevel\"] = 1\n",
    "        \n",
    "        df_combined[\"2FamilyConversion\"] = 0\n",
    "        df_combined.loc[np.logical_or(df_combined[\"PropertyType1\"] == \"2FamilyConversion\", df_combined[\"PropertyType2\"] == \"2FamilyConv\"), \"2FamilyConversion\"] = 1\n",
    "\n",
    "        df_combined[\"1Family\"] = 0\n",
    "        df_combined.loc[df_combined[\"PropertyType2\"] == \"1Family\", \"1Family\"] = 1\n",
    "\n",
    "        df_combined[\"Attic\"] = df_combined[\"PropertyType1\"].isin([\"1StoryAttic\"]).astype(int) \n",
    "        df_combined[\"Townhouse\"] = df_combined[\"PropertyType2\"].isin([\"Twnhs\", \"EndTownhouse\"]).astype(int) \n",
    "        df_combined[\"PUDMultilevel\"] = df_combined[\"PropertyType1\"].isin([\"MultilevelPlannedDevelopment\"]).astype(int) \n",
    "    \n",
    "        df_combined[\"Basement_Living_Area\"] = (df_combined[\"FinishedBasementType1\"].isin([\"GoodLivingQuarters\", \"AverageLivingQuarters\", \"BelowAverageLivingQuarters\"]).astype(int)*df_combined[\"FinishedBasementSF1\"]) + (df_combined[\"FinishedBasementType2\"].isin([\"GoodLivingQuarters\", \"AverageLivingQuarters\", \"BelowAverageLivingQuarters\"]).astype(int)*df_combined[\"FinishedBasementSF2\"])\n",
    "        df_combined[\"Basement_Rec_Area\"] = (df_combined[\"FinishedBasementType1\"].isin([\"RecRoom\"]).astype(int)*df_combined[\"FinishedBasementSF1\"]) + (df_combined[\"FinishedBasementType2\"].isin([\"RecRoom\"]).astype(int)*df_combined[\"FinishedBasementSF2\"])\n",
    "        df_combined[\"Basement_LQ_Area\"] = (df_combined[\"FinishedBasementType1\"].isin([\"LowQuality\"]).astype(int)*df_combined[\"FinishedBasementSF1\"]) + (df_combined[\"FinishedBasementType2\"].isin([\"LowQuality\"]).astype(int)*df_combined[\"FinishedBasementSF2\"])\n",
    "                \n",
    "        for materials in [\"AsbestosShingles\", \"AsphaltShingles\", \"CommonBrick\", \"FaceBrick\", \"CinderBlock\", \"Cement\", \"Hardboard\", \"ImitationStucco\", \"Metal\", \"Other\", \"Plywood\", \"Stone\", \"Stucco\", \"Vinyl\", \"WoodSiding\", \"WoodShingles\"]:\n",
    "            variable_name = \"Exterior_\" + materials + \"Proportion\"\n",
    "            df_combined[variable_name] = 0.5*(df_combined[\"Exterior1\"].isin([materials]).astype(int)) + .5*(df_combined[\"Exterior2\"].isin([materials]).astype(int))\n",
    "            \n",
    "        df_combined[\"Exterior_Unknown_Proportion\"] = df_combined[\"Exterior1\"].isna().astype(\"int64\")\n",
    "        \n",
    "        df_combined[\"Property_Type_1NA\"] = df_combined[\"PropertyType1\"].isna().astype(\"int64\")\n",
    "        \n",
    "    if drop_variables is False: \n",
    "        drop_variables_list = [] \n",
    "    elif drop_variables == \"comprehensive\":\n",
    "        drop_variables_list = [\"SaleType\", \"HeatingType\", \"Functional\", \"Utilities\", \"PropertyType1\", \"HouseStyle\", \"PropertyType2\", \"Neighborhood\", \"LotShape\", \"LandContour\", \"LandSlope\", \"PavedDriveType\", \"FenceQual\", \"StreetMaterial\", \"AlleyMaterial\", \"PoolQual\", \"GarageQual\", \"GarageCond\", \"FireplaceQual\", \"BsmtQual\", \"HeatingCond\", \"BsmtExposure\", \"BsmtCond\", \"ExteriorCond\", \"ExteriorQual\",\"FenceQual\", 'WoodDeckSF','OpenPorchSF', 'EnclosedPorchSF', '3SsnPorchSF', 'ScreenPorchSF', \"HalfBathCount\", \"BsmtHalfBathCount\", \"FullBathCount\", \"BsmtFullBathCount\", \"AlleyMaterial\", \"LocationAmenity1\", \"LocationAmenity2\", \"MoSold\", \"GarageYrBlt\", \"YearLastRemodeled\", \"YearBuilt\", \"SaleFlags\", \"Exterior1\", \"Exterior2\"]\n",
    "    elif drop_variables == \"minimal\":\n",
    "        drop_variables_list = [\"FenceQual\", \"Utilities\", \"PropertyType1\", \"HouseStyle\", \"PropertyType2\", \"Exterior1\", \"Exterior2\"]\n",
    "    else:\n",
    "        raise Exception(\"Invalid drop_variable input: options are False, comprehensive, and minimal.\")\n",
    "\n",
    "    if y_fld != \"PricePerArea\":               \n",
    "        y = df_combined[y_fld].values\n",
    "        drop_variables_list += [y_fld]\n",
    "    else:\n",
    "        y = np.log(df_combined.SalePrice/df_combined.LotArea).values\n",
    "        drop_variables_list += [\"SalePrice\"]\n",
    "        \n",
    "    df_combined.drop(drop_variables_list, axis=1, inplace=True)\n",
    "\n",
    "    if imputation_methodology != \"KNN\" and imputation_methodology != \"IterativeImputer\":\n",
    "        na_dict = fix_missing2(df_combined, imputation_split, imputation_methodology) \n",
    "    \n",
    "    if vars_to_cat == \"ordinal\":\n",
    "        for n in [\"LotShape\", \"LandSlope\", \"ExteriorQual\", \"ExteriorCond\", \"BsmtQual\", \"BsmtCond\", \n",
    "                  \"BsmtExposure\", \"FinishedBasementType1\", \"FinishedBasementType2\", \"HeatingCond\", \"KitchenQual\",\n",
    "                  \"Functional\", \"FireplaceQual\", \"GarageQual\", \"GarageQual\", \"GarageCond\", \"PoolQual\"]:\n",
    "            numericalize2(df_combined, n)\n",
    "    elif vars_to_cat == \"all\":\n",
    "        for n, _ in df_combined.items(): \n",
    "            numericalize2(df_combined, n)\n",
    "    elif vars_to_cat != \"none\":\n",
    "        raise Exception(\"Choose a valid categorization scheme!\")\n",
    "     \n",
    "    df_combined = pd.get_dummies(df_combined, dummy_na=True)\n",
    "    \n",
    "    df_combined.drop(list(df_combined.filter(regex='nan$',axis=1).loc[:, (df_combined == 0).all(axis=0)].columns), axis = 1, inplace = True)\n",
    "    df_combined.drop(list(df_combined.filter(regex='na$',axis=1).loc[:, (df_combined == 0).all(axis=0)].columns), axis = 1, inplace = True)\n",
    "    df_combined.drop(list(df_combined.loc[:, (df_combined == 0).all(axis=0)].columns), axis = 1, inplace = True)\n",
    "\n",
    "    if add_variables == True and drop_variables == \"minimal\":\n",
    "        print(\"Minimally dropping variables.\")\n",
    "        df_combined.drop([\"HeatingType_OtherWater\", \"VeneerArea_NA\", \"FenceQuality_None\", \"FenceType_None\", \"New\", \"Reg_Contract\",\n",
    "                              \"GarageQual_None\", \"GarageInterior_None\", \"GarageCond_None\", 'BsmtCond_None', 'BsmtExposure_NoBasement', 'Functional_ModerateDeduction', \"Functional_SevereDamage\",\n",
    "                              'BsmtQual_None', \"FenceType_None\", \"HeatingType_Floor\"], axis = 1, inplace = True, errors = \"ignore\")\n",
    "    elif add_variables == True and drop_variables == \"comprehensive\":\n",
    "        print(\"Comprehensively dropping variables.\")\n",
    "        df_combined.drop([\"VeneerArea_NA\", \"FenceQuality_None\", 'ZoningClass_C (all)', 'ZoningClass_FloatingVillage', 'ZoningClass_ResHighDens', 'ZoningClass_ResLowDens', 'ZoningClass_ResMedDens', \"GarageCarsCount_NA\", \"GarageInterior_None\", \"FenceType_None\"], axis = 1, inplace = True, errors = \"ignore\")        \n",
    "    elif drop_variables != False:\n",
    "        raise Exception(\"Invalid Entry for drop_variables!\")\n",
    "    else:\n",
    "        print(\"Not dropping variables.\")\n",
    "    \n",
    "    if log_size_variables is True:\n",
    "        size_variables = [\"AboveGradeAreaSF\", \"LotArea\"]\n",
    "        for var in size_variables:\n",
    "            new_variable_name = var + \"_logged\"\n",
    "            df_combined[new_variable_name] = np.log(df_combined[var].astype('float64'))\n",
    "        df_combined.drop(size_variables, axis=1, inplace=True)\n",
    "    \n",
    "    if square_variables is False: \n",
    "        variables_to_square = []\n",
    "    elif square_variables is True and log_size_variables is False:\n",
    "        variables_to_square = [\"HouseAge\", \"TotalBsmtSF\", \"AboveGradeAreaSF\", \"LotArea\", \"LotFrontage\", \"GarageArea\", \"1stFlrSF\"]\n",
    "    elif square_variables is True and log_size_variables is True:\n",
    "        variables_to_square = [\"HouseAge\", \"TotalBsmtSF\", \"AboveGradeAreaSF_logged\", \"LotArea_logged\", \"LotFrontage\", \"GarageArea\", \"1stFlrSF\"]\n",
    "    else:\n",
    "        raise Exception(\"Invalid Entry for Square_Variables!\")\n",
    "        \n",
    "    for var in variables_to_square:\n",
    "        new_variable_name = var + \"_sq\"\n",
    "        df_combined[new_variable_name] = df_combined[var]**2\n",
    "    \n",
    "    if df2 is None and n_validation is None and percentage_validation is None:\n",
    "        res = [df_combined, y, na_dict]\n",
    "    else:\n",
    "        df1, df2 = df_combined.iloc[:split_point, :], df_combined.iloc[split_point:, :]\n",
    "        y1, y2 = y[:split_point], y[split_point:]\n",
    "        res = [df1, y1, df2, y2, na_dict]\n",
    "\n",
    "    return res"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Read in data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [],
   "source": [
    "# Checks if Python interpreter is using folders that contain \"Kaggle\" as part of path\n",
    "# Don't put this in folder names!\n",
    "if \"kaggle\" in '\\t'.join(sys.path):\n",
    "    raw_train_input = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/train.csv')\n",
    "    raw_test_input = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/test.csv')\n",
    "    data_dictionary = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/data_description.txt', delimiter= ':', index_col=False, header = None, error_bad_lines=False)\n",
    "else:\n",
    "    raw_train_input = pd.read_csv('train.csv')\n",
    "    raw_test_input = pd.read_csv('test.csv')\n",
    "\n",
    "train = raw_train_input\n",
    "test = raw_test_input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "## AUTOMATICALLY PRODUCE VARIABLE DESCRIPTION\n",
    "\n",
    "if \"kaggle\" in '\\t'.join(sys.path):\n",
    "    data_dictionary = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/data_description.txt', delimiter= ':', index_col=False, header = None, error_bad_lines=False)\n",
    "    data_dictionary = data_dictionary[~pd.isna(data_dictionary[1])]\n",
    "    data_dictionary = data_dictionary[~data_dictionary[0].str.contains(\"\\t\")].T\n",
    "    data_dictionary.rename(columns=data_dictionary.iloc[0], inplace = True)\n",
    "    data_dictionary.drop([0], inplace = True)\n",
    "else:\n",
    "    data_dictionary = pd.read_csv('data_description.txt', delimiter= ':', index_col=False, header = None, error_bad_lines=False)\n",
    "    data_dictionary = data_dictionary[~pd.isna(data_dictionary[1])].T\n",
    "    data_dictionary.rename(columns=data_dictionary.iloc[0], inplace = True)\n",
    "    data_dictionary.drop([0], inplace = True)\n",
    "\n",
    "data_dictionary.rename(columns={'MSSubClass': 'PropertyType1', \"MSZoning\":\"ZoningClass\", \"Street\":\"StreetMaterial\", \n",
    "                       \"Alley\":\"AlleyMaterial\", \"Condition1\":\"LocationAmenity1\",\"Condition2\":\"LocationAmenity2\",\n",
    "                      \"BldgType\":\"PropertyType2\", \"YearRemodAdd\":\"YearLastRemodeled\", \"RoofMatl\":\"RoofMaterial\", \"Exterior1st\":\"Exterior1\", \"ExterCond\": \"ExteriorCond\", \"ExterQual\": \"ExteriorQual\",\n",
    "                      \"Exterior2nd\":\"Exterior2\", \"MasVnrType\":\"VeneerType\",\"MasVnrArea\":\"VeneerArea\",\n",
    "                      \"Foundation\":\"FoundationType\", \"Heating\": \"HeatingType\", \"HeatingQC\":\"HeatingCond\",\n",
    "                      \"Electrical\":\"ElectricalType\", \"GrLivArea\":\"AboveGradeAreaSF\", \"TotRmsAbvGrd\":\"AboveGradeRooms\", \"LowQualFinSF\": \"LowQualFinishedSF\",\n",
    "                      \"FireplaceQu\":\"FireplaceQual\", \"GarageFinish\":\"GarageInterior\", \"EnclosedPorch\":\"EnclosedPorchSF\", \"Bedroom\": \"BedroomAbvGrnd\", \n",
    "                      \"PoolArea\":\"PoolAreaSF\", \"3SsnPorch\":\"3SsnPorchSF\", \"ScreenPorch\":\"ScreenPorchSF\", \"Fence\":\"FenceQual\",\n",
    "                      \"SaleCondition\":\"SaleFlags\", 'GarageCars': 'GarageCarsCount', \"BsmtFinType1\": \"FinishedBasementType1\", \"BsmtFinType2\": \"FinishedBasementType2\", \n",
    "                      \"BsmtFinSF1\": \"FinishedBasementSF1\", \"BsmtFinSF2\": \"FinishedBasementSF2\", \"BsmtUnfSF\":\"UnfinishedBasementSF\", \n",
    "                      \"FullBath\":\"FullBathCount\", \"HalfBath\": \"HalfBathCount\", \"BsmtFullBath\": \"BsmtFullBathCount\", \"BsmtHalfBath\": \"BsmtHalfBathCount\", \n",
    "                      \"Fireplaces\": \"FireplaceCount\", \"KitchenAbvGr\": \"KitchenCount\", \"PoolQC\": \"PoolQual\", \"PavedDrive\":\"PavedDriveType\"}, inplace=True)\n",
    "\n",
    "data_dictionary = data_dictionary.T.rename(columns = {1: \"Definition\"})\n",
    "data_dictionary.index.names = [None] \n",
    "data_dictionary.reset_index(inplace = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## EDA and preprocessing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'SalePrice'}\n",
      "Train: YrSold\n",
      "2006    314\n",
      "2007    329\n",
      "2008    304\n",
      "2009    338\n",
      "2010    175\n",
      "Name: Id, dtype: int64\n",
      "Test: YrSold\n",
      "2006    305\n",
      "2007    363\n",
      "2008    318\n",
      "2009    309\n",
      "2010    164\n",
      "Name: Id, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "# Check column matches - only difference between train and test is SalePrice = label, time period is shared\n",
    "\n",
    "print(set(train.columns) - set(test.columns))\n",
    "print(\"Train:\", train.groupby('YrSold')[\"Id\"].count())\n",
    "print(\"Test:\", test.groupby('YrSold')[\"Id\"].count())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>count</th>\n",
       "      <th>mean</th>\n",
       "      <th>std</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Id</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>730.500000</td>\n",
       "      <td>421.610009</td>\n",
       "      <td>1.0</td>\n",
       "      <td>365.75</td>\n",
       "      <td>730.5</td>\n",
       "      <td>1095.25</td>\n",
       "      <td>1460.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MSSubClass</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>56.897260</td>\n",
       "      <td>42.300571</td>\n",
       "      <td>20.0</td>\n",
       "      <td>20.00</td>\n",
       "      <td>50.0</td>\n",
       "      <td>70.00</td>\n",
       "      <td>190.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LotFrontage</th>\n",
       "      <td>1201.0</td>\n",
       "      <td>70.049958</td>\n",
       "      <td>24.284752</td>\n",
       "      <td>21.0</td>\n",
       "      <td>59.00</td>\n",
       "      <td>69.0</td>\n",
       "      <td>80.00</td>\n",
       "      <td>313.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LotArea</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>10516.828082</td>\n",
       "      <td>9981.264932</td>\n",
       "      <td>1300.0</td>\n",
       "      <td>7553.50</td>\n",
       "      <td>9478.5</td>\n",
       "      <td>11601.50</td>\n",
       "      <td>215245.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>OverallQual</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>6.099315</td>\n",
       "      <td>1.382997</td>\n",
       "      <td>1.0</td>\n",
       "      <td>5.00</td>\n",
       "      <td>6.0</td>\n",
       "      <td>7.00</td>\n",
       "      <td>10.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>OverallCond</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>5.575342</td>\n",
       "      <td>1.112799</td>\n",
       "      <td>1.0</td>\n",
       "      <td>5.00</td>\n",
       "      <td>5.0</td>\n",
       "      <td>6.00</td>\n",
       "      <td>9.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YearBuilt</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1971.267808</td>\n",
       "      <td>30.202904</td>\n",
       "      <td>1872.0</td>\n",
       "      <td>1954.00</td>\n",
       "      <td>1973.0</td>\n",
       "      <td>2000.00</td>\n",
       "      <td>2010.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YearRemodAdd</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1984.865753</td>\n",
       "      <td>20.645407</td>\n",
       "      <td>1950.0</td>\n",
       "      <td>1967.00</td>\n",
       "      <td>1994.0</td>\n",
       "      <td>2004.00</td>\n",
       "      <td>2010.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MasVnrArea</th>\n",
       "      <td>1452.0</td>\n",
       "      <td>103.685262</td>\n",
       "      <td>181.066207</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>166.00</td>\n",
       "      <td>1600.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BsmtFinSF1</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>443.639726</td>\n",
       "      <td>456.098091</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>383.5</td>\n",
       "      <td>712.25</td>\n",
       "      <td>5644.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BsmtFinSF2</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>46.549315</td>\n",
       "      <td>161.319273</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1474.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BsmtUnfSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>567.240411</td>\n",
       "      <td>441.866955</td>\n",
       "      <td>0.0</td>\n",
       "      <td>223.00</td>\n",
       "      <td>477.5</td>\n",
       "      <td>808.00</td>\n",
       "      <td>2336.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TotalBsmtSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1057.429452</td>\n",
       "      <td>438.705324</td>\n",
       "      <td>0.0</td>\n",
       "      <td>795.75</td>\n",
       "      <td>991.5</td>\n",
       "      <td>1298.25</td>\n",
       "      <td>6110.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1stFlrSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1162.626712</td>\n",
       "      <td>386.587738</td>\n",
       "      <td>334.0</td>\n",
       "      <td>882.00</td>\n",
       "      <td>1087.0</td>\n",
       "      <td>1391.25</td>\n",
       "      <td>4692.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2ndFlrSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>346.992466</td>\n",
       "      <td>436.528436</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>728.00</td>\n",
       "      <td>2065.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LowQualFinSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>5.844521</td>\n",
       "      <td>48.623081</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>572.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GrLivArea</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1515.463699</td>\n",
       "      <td>525.480383</td>\n",
       "      <td>334.0</td>\n",
       "      <td>1129.50</td>\n",
       "      <td>1464.0</td>\n",
       "      <td>1776.75</td>\n",
       "      <td>5642.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BsmtFullBath</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>0.425342</td>\n",
       "      <td>0.518911</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BsmtHalfBath</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>0.057534</td>\n",
       "      <td>0.238753</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FullBath</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1.565068</td>\n",
       "      <td>0.550916</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.00</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>HalfBath</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>0.382877</td>\n",
       "      <td>0.502885</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BedroomAbvGr</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>2.866438</td>\n",
       "      <td>0.815778</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2.00</td>\n",
       "      <td>3.0</td>\n",
       "      <td>3.00</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>KitchenAbvGr</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1.046575</td>\n",
       "      <td>0.220338</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TotRmsAbvGrd</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>6.517808</td>\n",
       "      <td>1.625393</td>\n",
       "      <td>2.0</td>\n",
       "      <td>5.00</td>\n",
       "      <td>6.0</td>\n",
       "      <td>7.00</td>\n",
       "      <td>14.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fireplaces</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>0.613014</td>\n",
       "      <td>0.644666</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GarageYrBlt</th>\n",
       "      <td>1379.0</td>\n",
       "      <td>1978.506164</td>\n",
       "      <td>24.689725</td>\n",
       "      <td>1900.0</td>\n",
       "      <td>1961.00</td>\n",
       "      <td>1980.0</td>\n",
       "      <td>2002.00</td>\n",
       "      <td>2010.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GarageCars</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>1.767123</td>\n",
       "      <td>0.747315</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.00</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GarageArea</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>472.980137</td>\n",
       "      <td>213.804841</td>\n",
       "      <td>0.0</td>\n",
       "      <td>334.50</td>\n",
       "      <td>480.0</td>\n",
       "      <td>576.00</td>\n",
       "      <td>1418.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>WoodDeckSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>94.244521</td>\n",
       "      <td>125.338794</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>168.00</td>\n",
       "      <td>857.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>OpenPorchSF</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>46.660274</td>\n",
       "      <td>66.256028</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>25.0</td>\n",
       "      <td>68.00</td>\n",
       "      <td>547.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>EnclosedPorch</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>21.954110</td>\n",
       "      <td>61.119149</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>552.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3SsnPorch</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>3.409589</td>\n",
       "      <td>29.317331</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>508.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ScreenPorch</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>15.060959</td>\n",
       "      <td>55.757415</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>480.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PoolArea</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>2.758904</td>\n",
       "      <td>40.177307</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>738.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MiscVal</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>43.489041</td>\n",
       "      <td>496.123024</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>15500.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MoSold</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>6.321918</td>\n",
       "      <td>2.703626</td>\n",
       "      <td>1.0</td>\n",
       "      <td>5.00</td>\n",
       "      <td>6.0</td>\n",
       "      <td>8.00</td>\n",
       "      <td>12.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YrSold</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>2007.815753</td>\n",
       "      <td>1.328095</td>\n",
       "      <td>2006.0</td>\n",
       "      <td>2007.00</td>\n",
       "      <td>2008.0</td>\n",
       "      <td>2009.00</td>\n",
       "      <td>2010.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SalePrice</th>\n",
       "      <td>1460.0</td>\n",
       "      <td>180921.195890</td>\n",
       "      <td>79442.502883</td>\n",
       "      <td>34900.0</td>\n",
       "      <td>129975.00</td>\n",
       "      <td>163000.0</td>\n",
       "      <td>214000.00</td>\n",
       "      <td>755000.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                count           mean           std      min        25%  \\\n",
       "Id             1460.0     730.500000    421.610009      1.0     365.75   \n",
       "MSSubClass     1460.0      56.897260     42.300571     20.0      20.00   \n",
       "LotFrontage    1201.0      70.049958     24.284752     21.0      59.00   \n",
       "LotArea        1460.0   10516.828082   9981.264932   1300.0    7553.50   \n",
       "OverallQual    1460.0       6.099315      1.382997      1.0       5.00   \n",
       "OverallCond    1460.0       5.575342      1.112799      1.0       5.00   \n",
       "YearBuilt      1460.0    1971.267808     30.202904   1872.0    1954.00   \n",
       "YearRemodAdd   1460.0    1984.865753     20.645407   1950.0    1967.00   \n",
       "MasVnrArea     1452.0     103.685262    181.066207      0.0       0.00   \n",
       "BsmtFinSF1     1460.0     443.639726    456.098091      0.0       0.00   \n",
       "BsmtFinSF2     1460.0      46.549315    161.319273      0.0       0.00   \n",
       "BsmtUnfSF      1460.0     567.240411    441.866955      0.0     223.00   \n",
       "TotalBsmtSF    1460.0    1057.429452    438.705324      0.0     795.75   \n",
       "1stFlrSF       1460.0    1162.626712    386.587738    334.0     882.00   \n",
       "2ndFlrSF       1460.0     346.992466    436.528436      0.0       0.00   \n",
       "LowQualFinSF   1460.0       5.844521     48.623081      0.0       0.00   \n",
       "GrLivArea      1460.0    1515.463699    525.480383    334.0    1129.50   \n",
       "BsmtFullBath   1460.0       0.425342      0.518911      0.0       0.00   \n",
       "BsmtHalfBath   1460.0       0.057534      0.238753      0.0       0.00   \n",
       "FullBath       1460.0       1.565068      0.550916      0.0       1.00   \n",
       "HalfBath       1460.0       0.382877      0.502885      0.0       0.00   \n",
       "BedroomAbvGr   1460.0       2.866438      0.815778      0.0       2.00   \n",
       "KitchenAbvGr   1460.0       1.046575      0.220338      0.0       1.00   \n",
       "TotRmsAbvGrd   1460.0       6.517808      1.625393      2.0       5.00   \n",
       "Fireplaces     1460.0       0.613014      0.644666      0.0       0.00   \n",
       "GarageYrBlt    1379.0    1978.506164     24.689725   1900.0    1961.00   \n",
       "GarageCars     1460.0       1.767123      0.747315      0.0       1.00   \n",
       "GarageArea     1460.0     472.980137    213.804841      0.0     334.50   \n",
       "WoodDeckSF     1460.0      94.244521    125.338794      0.0       0.00   \n",
       "OpenPorchSF    1460.0      46.660274     66.256028      0.0       0.00   \n",
       "EnclosedPorch  1460.0      21.954110     61.119149      0.0       0.00   \n",
       "3SsnPorch      1460.0       3.409589     29.317331      0.0       0.00   \n",
       "ScreenPorch    1460.0      15.060959     55.757415      0.0       0.00   \n",
       "PoolArea       1460.0       2.758904     40.177307      0.0       0.00   \n",
       "MiscVal        1460.0      43.489041    496.123024      0.0       0.00   \n",
       "MoSold         1460.0       6.321918      2.703626      1.0       5.00   \n",
       "YrSold         1460.0    2007.815753      1.328095   2006.0    2007.00   \n",
       "SalePrice      1460.0  180921.195890  79442.502883  34900.0  129975.00   \n",
       "\n",
       "                    50%        75%       max  \n",
       "Id                730.5    1095.25    1460.0  \n",
       "MSSubClass         50.0      70.00     190.0  \n",
       "LotFrontage        69.0      80.00     313.0  \n",
       "LotArea          9478.5   11601.50  215245.0  \n",
       "OverallQual         6.0       7.00      10.0  \n",
       "OverallCond         5.0       6.00       9.0  \n",
       "YearBuilt        1973.0    2000.00    2010.0  \n",
       "YearRemodAdd     1994.0    2004.00    2010.0  \n",
       "MasVnrArea          0.0     166.00    1600.0  \n",
       "BsmtFinSF1        383.5     712.25    5644.0  \n",
       "BsmtFinSF2          0.0       0.00    1474.0  \n",
       "BsmtUnfSF         477.5     808.00    2336.0  \n",
       "TotalBsmtSF       991.5    1298.25    6110.0  \n",
       "1stFlrSF         1087.0    1391.25    4692.0  \n",
       "2ndFlrSF            0.0     728.00    2065.0  \n",
       "LowQualFinSF        0.0       0.00     572.0  \n",
       "GrLivArea        1464.0    1776.75    5642.0  \n",
       "BsmtFullBath        0.0       1.00       3.0  \n",
       "BsmtHalfBath        0.0       0.00       2.0  \n",
       "FullBath            2.0       2.00       3.0  \n",
       "HalfBath            0.0       1.00       2.0  \n",
       "BedroomAbvGr        3.0       3.00       8.0  \n",
       "KitchenAbvGr        1.0       1.00       3.0  \n",
       "TotRmsAbvGrd        6.0       7.00      14.0  \n",
       "Fireplaces          1.0       1.00       3.0  \n",
       "GarageYrBlt      1980.0    2002.00    2010.0  \n",
       "GarageCars          2.0       2.00       4.0  \n",
       "GarageArea        480.0     576.00    1418.0  \n",
       "WoodDeckSF          0.0     168.00     857.0  \n",
       "OpenPorchSF        25.0      68.00     547.0  \n",
       "EnclosedPorch       0.0       0.00     552.0  \n",
       "3SsnPorch           0.0       0.00     508.0  \n",
       "ScreenPorch         0.0       0.00     480.0  \n",
       "PoolArea            0.0       0.00     738.0  \n",
       "MiscVal             0.0       0.00   15500.0  \n",
       "MoSold              6.0       8.00      12.0  \n",
       "YrSold           2008.0    2009.00    2010.0  \n",
       "SalePrice      163000.0  214000.00  755000.0  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.set_option('display.max_rows', 100)\n",
    "train.describe().T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Basic data cleansing and conversion of target to log, create three new age variables\n",
    "\n",
    "if \"Id\" in train:\n",
    "    train = train.set_index('Id')\n",
    "    \n",
    "if \"Id\" in test:\n",
    "    test = test.set_index('Id')\n",
    "\n",
    "train['PricePerArea'] = np.log(train.SalePrice/train.LotArea)\n",
    "train.SalePrice = np.log(train.SalePrice)\n",
    "\n",
    "train_int = train.copy()\n",
    "train = train.drop(['PricePerArea'],axis=1)\n",
    "train2 = train_int.drop(['SalePrice'],axis=1)\n",
    "\n",
    "for dataset_names in (train, test):\n",
    "    dataset_names.index.names = [None] \n",
    "    dataset_names[\"HouseAge\"] = dataset_names.YrSold - dataset_names.YearBuilt \n",
    "    dataset_names[\"RemodAge\"] = dataset_names.YrSold - dataset_names.YearRemodAdd \n",
    "    dataset_names[\"GarageAge\"] = dataset_names.YrSold - dataset_names.GarageYrBlt\n",
    "    \n",
    "    col_list = [col for col in train.columns if 'Bsmt' in col and 'SF' in col]\n",
    "    for variable_names in col_list:\n",
    "        dataset_names[variable_names] = dataset_names[variable_names].astype(\"float64\")\n",
    "        \n",
    "    dataset_names[\"BsmtFullBath\"] = dataset_names[\"BsmtFullBath\"].astype(\"float64\")\n",
    "    dataset_names[\"BsmtHalfBath\"] = dataset_names[\"BsmtHalfBath\"].astype(\"float64\")\n",
    "    dataset_names[\"GarageCars\"] = dataset_names[\"GarageCars\"].astype(\"float64\")\n",
    "    dataset_names[\"GarageArea\"] = dataset_names[\"GarageArea\"].astype(\"float64\")\n",
    "    \n",
    "    dataset_names[\"MSSubClass\"] = dataset_names[\"MSSubClass\"].astype(str)\n",
    "    \n",
    "    for variable_names in [\"Alley\", \"BsmtQual\", \"BsmtCond\", \"BsmtExposure\", \"BsmtFinType1\", \"BsmtFinType2\", \"FireplaceQu\", \"GarageType\", \"GarageFinish\", \"GarageQual\", \"GarageCond\", \"PoolQC\", \"Fence\", \"MiscFeature\"]:\n",
    "        dataset_names.loc[pd.isna(dataset_names[variable_names]), variable_names] = \"None\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LotFrontage    259\n",
      "MasVnrType       8\n",
      "MasVnrArea       8\n",
      "Electrical       1\n",
      "GarageYrBlt     81\n",
      "GarageAge       81\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "# Investigate data gaps\n",
    "list_of_nas = train.isna().sum()\n",
    "list_of_nas = list_of_nas[~(list_of_nas == 0)]\n",
    "print(list_of_nas)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#for column_names in test.columns:\n",
    "#    if pvalue_sample_differences(test, train, column_names) < .15:\n",
    "#        print(column_names + \": \" + str(pvalue_sample_differences(test, train, column_names)))\n",
    "#        print(\"TRAIN:\")\n",
    "#        print(test[column_names].describe())\n",
    "#        print(\"TEST:\")\n",
    "#        print(train[column_names].describe())\n",
    "#        \n",
    "#print(test.shape[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LotFrontage\n",
      "935     313.0\n",
      "1299    313.0\n",
      "Name: LotFrontage, dtype: float64\n",
      "\n",
      "\n",
      "LotArea\n",
      "250    159000\n",
      "314    215245\n",
      "336    164660\n",
      "707    115149\n",
      "Name: LotArea, dtype: int64\n",
      "\n",
      "\n",
      "Utilities\n",
      "NoSeWa    1\n",
      "Name: Utilities, dtype: int64\n",
      "\n",
      "\n",
      "Neighborhood\n",
      "Blueste    2\n",
      "Name: Neighborhood, dtype: int64\n",
      "\n",
      "\n",
      "Condition1\n",
      "RRNe    2\n",
      "Name: Condition1, dtype: int64\n",
      "\n",
      "\n",
      "Condition2\n",
      "PosN      2\n",
      "RRNn      2\n",
      "Artery    2\n",
      "RRAe      1\n",
      "RRAn      1\n",
      "PosA      1\n",
      "Name: Condition2, dtype: int64\n",
      "\n",
      "\n",
      "RoofStyle\n",
      "Shed    2\n",
      "Name: RoofStyle, dtype: int64\n",
      "\n",
      "\n",
      "RoofMatl\n",
      "Membran    1\n",
      "ClyTile    1\n",
      "Metal      1\n",
      "Roll       1\n",
      "Name: RoofMatl, dtype: int64\n",
      "\n",
      "\n",
      "Exterior1st\n",
      "Stone      2\n",
      "BrkComm    2\n",
      "AsphShn    1\n",
      "ImStucc    1\n",
      "CBlock     1\n",
      "Name: Exterior1st, dtype: int64\n",
      "\n",
      "\n",
      "Exterior2nd\n",
      "AsphShn    3\n",
      "Other      1\n",
      "CBlock     1\n",
      "Name: Exterior2nd, dtype: int64\n",
      "\n",
      "\n",
      "ExterCond\n",
      "Ex    3\n",
      "Po    1\n",
      "Name: ExterCond, dtype: int64\n",
      "\n",
      "\n",
      "Foundation\n",
      "Wood    3\n",
      "Name: Foundation, dtype: int64\n",
      "\n",
      "\n",
      "BsmtCond\n",
      "Po    2\n",
      "Name: BsmtCond, dtype: int64\n",
      "\n",
      "\n",
      "BsmtFinSF1\n",
      "1299    5644.0\n",
      "Name: BsmtFinSF1, dtype: float64\n",
      "\n",
      "\n",
      "TotalBsmtSF\n",
      "1299    6110.0\n",
      "Name: TotalBsmtSF, dtype: float64\n",
      "\n",
      "\n",
      "Heating\n",
      "OthW     2\n",
      "Floor    1\n",
      "Name: Heating, dtype: int64\n",
      "\n",
      "\n",
      "HeatingQC\n",
      "Po    1\n",
      "Name: HeatingQC, dtype: int64\n",
      "\n",
      "\n",
      "Electrical\n",
      "FuseP    3\n",
      "Mix      1\n",
      "Name: Electrical, dtype: int64\n",
      "\n",
      "\n",
      "1stFlrSF\n",
      "1299    4692\n",
      "Name: 1stFlrSF, dtype: int64\n",
      "\n",
      "\n",
      "KitchenAbvGr\n",
      "49     3\n",
      "810    3\n",
      "Name: KitchenAbvGr, dtype: int64\n",
      "\n",
      "\n",
      "Functional\n",
      "Sev    1\n",
      "Name: Functional, dtype: int64\n",
      "\n",
      "\n",
      "GarageQual\n",
      "Po    3\n",
      "Ex    3\n",
      "Name: GarageQual, dtype: int64\n",
      "\n",
      "\n",
      "GarageCond\n",
      "Ex    2\n",
      "Name: GarageCond, dtype: int64\n",
      "\n",
      "\n",
      "3SsnPorch\n",
      "Series([], Name: 3SsnPorch, dtype: int64)\n",
      "\n",
      "\n",
      "PoolQC\n",
      "Gd    3\n",
      "Ex    2\n",
      "Fa    2\n",
      "Name: PoolQC, dtype: int64\n",
      "\n",
      "\n",
      "MiscFeature\n",
      "Gar2    2\n",
      "Othr    2\n",
      "TenC    1\n",
      "Name: MiscFeature, dtype: int64\n",
      "\n",
      "\n",
      "MiscVal\n",
      "Series([], Name: MiscVal, dtype: int64)\n",
      "\n",
      "\n",
      "SaleType\n",
      "Oth    3\n",
      "Con    2\n",
      "Name: SaleType, dtype: int64\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# List outliers\n",
    "#for numeric > 10std away from mean\n",
    "#<4 categories\n",
    "\n",
    "for column_names in train.columns:\n",
    "    if train[train[column_names].astype(bool)][column_names].dtype == \"int64\" or train[train[column_names].astype(bool)][column_names].dtype == \"float64\":\n",
    "        if len(train[train[column_names].astype(bool)].loc[train[train[column_names].astype(bool)][column_names] >= train[train[column_names].astype(bool)][column_names].mean() + 8*(train[column_names].std()), column_names]) > 0:\n",
    "            print(column_names)\n",
    "            print(train[train[column_names].astype(bool)].loc[train[train[column_names].astype(bool)][column_names] >= train[train[column_names].astype(bool)][column_names].mean() + 8*(train[train[column_names].astype(bool)][column_names].std()), column_names])\n",
    "            print(\"\\n\")\n",
    "    if train[column_names].dtype == \"object\":\n",
    "        list_of_counts = train[column_names].value_counts()\n",
    "        list_of_counts = list_of_counts[list_of_counts < 4]\n",
    "        if len(list_of_counts) > 0:\n",
    "            print(column_names)\n",
    "            print(list_of_counts)\n",
    "            print(\"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LotFrontage (Normality): 2.003834562954084e-29\n",
      "LotArea (Normality): 0.0\n",
      "OverallQual (Normality): 2.6899265770757965e-22\n",
      "OverallCond (Normality): 6.778475726002203e-37\n",
      "YearBuilt (Normality): 2.7700502478701884e-26\n",
      "YearRemodAdd (Normality): 6.7254540933404426e-34\n",
      "MasVnrArea (Normality): 0.0\n",
      "BsmtFinSF1 (Normality): 2.8140120808220724e-35\n",
      "BsmtFinSF2 (Normality): 0.0\n",
      "BsmtUnfSF (Normality): 1.6411276150348104e-25\n",
      "TotalBsmtSF (Normality): 1.6116577553235961e-27\n",
      "1stFlrSF (Normality): 4.5177059476948354e-26\n",
      "2ndFlrSF (Normality): 2.5150504837701817e-41\n",
      "LowQualFinSF (Normality): 0.0\n",
      "GrLivArea (Normality): 6.598091159538852e-26\n",
      "BsmtFullBath (Normality): 0.0\n",
      "BsmtHalfBath (Normality): 0.0\n",
      "FullBath (Normality): 4.203895392974451e-44\n",
      "HalfBath (Normality): 0.0\n",
      "BedroomAbvGr (Normality): 4.1095632627576274e-35\n",
      "KitchenAbvGr (Normality): 0.0\n",
      "TotRmsAbvGrd (Normality): 2.0049259626764796e-23\n",
      "Fireplaces (Normality): 4.833078403456294e-42\n",
      "GarageYrBlt (Normality): 2.815365684643269e-26\n",
      "GarageCars (Normality): 2.305242068923655e-36\n",
      "GarageArea (Normality): 4.018698690337087e-15\n",
      "WoodDeckSF (Normality): 3.2283114021115136e-41\n",
      "OpenPorchSF (Normality): 1.1350517561031018e-43\n",
      "EnclosedPorch (Normality): 0.0\n",
      "3SsnPorch (Normality): 0.0\n",
      "ScreenPorch (Normality): 0.0\n",
      "PoolArea (Normality): 0.0\n",
      "MiscVal (Normality): 0.0\n",
      "MoSold (Normality): 3.190920019583739e-17\n",
      "YrSold (Normality): 3.4190127761463244e-30\n",
      "SalePrice (Normality): 1.1490678986092462e-07\n",
      "HouseAge (Normality): 3.203471075167616e-26\n",
      "RemodAge (Normality): 1.8268789332961113e-33\n",
      "GarageAge (Normality): 2.6422547812243845e-26\n"
     ]
    }
   ],
   "source": [
    "# Test for normality, generally right-skewed\n",
    "\n",
    "for column_names in train.columns:\n",
    "    if train[column_names].dtype == \"int64\" or train[column_names].dtype == \"float64\":\n",
    "        print(column_names + \" (Normality): \" + str(scipy.stats.shapiro(train[column_names].dropna())[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Rename variables\n",
    "\n",
    "train_cats(train)\n",
    "apply_cats(test, train)\n",
    "\n",
    "\n",
    "for dataset_names in (train, test):\n",
    "    dataset_names.LotShape.cat.set_categories([\"Reg\",\"IR1\",\"IR2\",\"IR3\"], ordered = True, inplace = True)\n",
    "    \n",
    "    dataset_names.Utilities.cat.set_categories([\"AllPub\",\"NoSewr\",\"NoSeWa\",\"ELO\"], ordered = True, inplace = True)\n",
    "    dataset_names.Utilities.cat.rename_categories({\"AllPub\": 'All', \"NoSewr\": \"NoSewer\", \"NoSeWa\": \"NoSewerWater\", \"ELO\": \"OnlyElectric\"}, inplace = True)\n",
    "    \n",
    "    dataset_names.LandSlope.cat.set_categories([\"Gtl\",\"Mod\",\"Sev\"], ordered = True, inplace = True)\n",
    "    dataset_names.LandSlope.cat.rename_categories({\"Gtl\": 'Gentle', \"Mod\": \"Moderate\", \"Sev\": \"Severe\"}, inplace = True)\n",
    "    \n",
    "    dataset_names.BsmtExposure.cat.set_categories([\"Gd\",\"Av\",\"Mn\",\"No\",\"None\"], ordered = True, inplace = True)\n",
    "    dataset_names.BsmtExposure.cat.rename_categories({\"Gd\": 'Good', \"Av\": \"Average\", \"Mn\": \"Fair\", \"No\": \"Poor\", \"None\": \"NoBasement\"}, inplace = True)\n",
    "    \n",
    "    dataset_names.GarageFinish.cat.set_categories([\"Fin\",\"RFn\",\"Unf\",\"None\"], ordered = True, inplace = True)\n",
    "    dataset_names.GarageFinish.cat.rename_categories({\"Fin\": 'Finished', \"RFn\": \"RoughFinished\", \"Unf\": \"Unfinished\"}, inplace = True)\n",
    "    \n",
    "    dataset_names.PoolQC.cat.set_categories([\"Ex\",\"Gd\",\"TA\",\"Fa\",\"None\"], ordered = True, inplace = True)\n",
    "    dataset_names.PoolQC.cat.rename_categories({'Ex': 'Excellent', 'Gd': 'Good', 'TA': 'Average', 'FA': 'Fair'}, inplace = True)\n",
    "    \n",
    "    for variable_names in [\"ExterQual\", \"ExterCond\", \"KitchenQual\", \"HeatingQC\"]:\n",
    "        dataset_names[variable_names].cat.set_categories([\"Ex\",\"Gd\",\"TA\",\"Fa\",\"Po\"], ordered = True, inplace = True)\n",
    "        dataset_names[variable_names].cat.rename_categories({'Ex': 'Excellent', 'Gd': 'Good', 'TA': 'Average', 'FA': 'Fair', 'Po': 'Poor'}, inplace = True)\n",
    "\n",
    "    for variable_names in [\"BsmtFinType1\", \"BsmtFinType2\"]:\n",
    "        dataset_names[variable_names].cat.set_categories([\"GLQ\",\"ALQ\",\"BLQ\",\"Rec\",\"LwQ\",\"Unf\", \"None\"], ordered = True, inplace = True)\n",
    "        dataset_names[variable_names].cat.rename_categories({'GLQ': 'GoodLivingQuarters', 'ALQ': 'AverageLivingQuarters', 'BLQ': 'BelowAverageLivingQuarters', 'Rec': 'RecRoom', 'LwQ': 'LowQuality', 'Unf': 'Unfinished'}, inplace = True)\n",
    "\n",
    "    for variable_names in [\"BsmtQual\", \"BsmtCond\", \"FireplaceQu\", \"GarageQual\", \"GarageCond\"]:\n",
    "        dataset_names[variable_names].cat.set_categories([\"Ex\",\"Gd\",\"TA\",\"Fa\",\"Po\",\"None\"], ordered = True, inplace = True)\n",
    "        dataset_names[variable_names].cat.rename_categories({'Ex': 'Excellent', 'Gd': 'Good', 'TA': 'Average', 'FA': 'Fair', 'Po': 'Poor'}, inplace = True)\n",
    "        \n",
    "    dataset_names.rename(columns={'MSSubClass': 'PropertyType1', \"MSZoning\":\"ZoningClass\", \"Street\":\"StreetMaterial\", \n",
    "                       \"Alley\":\"AlleyMaterial\", \"Condition1\":\"LocationAmenity1\",\"Condition2\":\"LocationAmenity2\",\n",
    "                      \"BldgType\":\"PropertyType2\", \"YearRemodAdd\":\"YearLastRemodeled\", \"RoofMatl\":\"RoofMaterial\", \"Exterior1st\":\"Exterior1\", \"ExterCond\": \"ExteriorCond\", \"ExterQual\": \"ExteriorQual\",\n",
    "                      \"Exterior2nd\":\"Exterior2\", \"MasVnrType\":\"VeneerType\",\"MasVnrArea\":\"VeneerArea\",\n",
    "                      \"Foundation\":\"FoundationType\", \"Heating\": \"HeatingType\", \"HeatingQC\":\"HeatingCond\",\n",
    "                      \"Electrical\":\"ElectricalType\", \"GrLivArea\":\"AboveGradeAreaSF\", \"TotRmsAbvGrd\":\"AboveGradeRooms\", \"LowQualFinSF\": \"LowQualFinishedSF\",\n",
    "                      \"FireplaceQu\":\"FireplaceQual\", \"GarageFinish\":\"GarageInterior\", \"EnclosedPorch\":\"EnclosedPorchSF\", \"Bedroom\": \"BedroomAbvGrnd\", \n",
    "                      \"PoolArea\":\"PoolAreaSF\", \"3SsnPorch\":\"3SsnPorchSF\", \"ScreenPorch\":\"ScreenPorchSF\", \"Fence\":\"FenceQual\",\n",
    "                      \"SaleCondition\":\"SaleFlags\", 'GarageCars': 'GarageCarsCount', \"BsmtFinType1\": \"FinishedBasementType1\", \"BsmtFinType2\": \"FinishedBasementType2\", \n",
    "                      \"BsmtFinSF1\": \"FinishedBasementSF1\", \"BsmtFinSF2\": \"FinishedBasementSF2\", \"BsmtUnfSF\":\"UnfinishedBasementSF\", \n",
    "                      \"FullBath\":\"FullBathCount\", \"HalfBath\": \"HalfBathCount\", \"BsmtFullBath\": \"BsmtFullBathCount\", \"BsmtHalfBath\": \"BsmtHalfBathCount\", \n",
    "                      \"Fireplaces\": \"FireplaceCount\", \"KitchenAbvGr\": \"KitchenCount\", \"PoolQC\": \"PoolQual\", \"PavedDrive\":\"PavedDriveType\"}, inplace=True)\n",
    "    \n",
    "    dataset_names.Neighborhood.cat.rename_categories({'Timber': 'Timberland', 'StoneBr': 'StoneBrook',\n",
    "                                                     'Somerst': 'Somerset', 'SawyerW':'SawyerWest',\n",
    "                                                     'Crawfor': 'Crawford', 'Blueste': 'Bluestem',\n",
    "                                                     'Mitchel': 'Mitchell', 'Names': 'NorthAmes',\n",
    "                                                     'ClearCr': 'ClearCreek', 'Blmngtn': 'BloomingtonHeights',\n",
    "                                                     'BrDale': 'Briardale', 'BrkSide': 'Brookside',\n",
    "                                                     'CollgCr': 'CollegeCreek', 'MeadowV': 'MeadowVillage',\n",
    "                                                     'NoRidge': 'Northridge', 'NPkVill': 'NorthparkVilla', 'NridgHt': 'NorthridgeHeights'}, inplace = True)\n",
    "    \n",
    "    dataset_names.PropertyType1.cat.rename_categories({\"20\": \"1StoryPostWar\", \"30\": \"1StoryPreWar\", \"40\": \"1StoryAttic\", \"45\": \"1.5StoryUnfinished\", \"50\": \"1.5StoryFinished\", \n",
    "                                                      \"60\": \"2StoryPostWar\", \"70\": \"2StoryPreWar\", \"75\": \"2.5Story\", \"80\": \"SplitNoFoyer\", \"85\": \"SplitFoyer\", \"90\": \"Duplex\", \n",
    "                                                       \"120\": \"1StoryPlannedDevelopment\", \"150\": \"1.5StoryPlannedDevelopment\", \"160\": \"2StoryPlannedDevelopment\", \"180\": \"MultilevelPlannedDevelopment\",\n",
    "                                                      \"190\": \"2FamilyConversion\"}, inplace = True)\n",
    "    \n",
    "    for variable_names in [\"LocationAmenity1\", \"LocationAmenity2\"]:\n",
    "        dataset_names[variable_names].cat.rename_categories({\"Artery\": \"ArteryStreet\", \"Feedr\": \"FeederStreet\", \"Norm\": \"None\", \"RRNn\": \"CloseNSRail\", \n",
    "                                                             \"RRNe\": \"CloseEWRail\", \"RRAn\": \"AdjacentNSRail\", \"RRAe\": \"AdjacentEWRail\", \"PosA\": \"AdjacentAmenity\", \"PosN\": \"CloseAmenity\"}, inplace = True)\n",
    "        \n",
    "    for variable_names in [\"Exterior1\", \"Exterior2\"]:\n",
    "        dataset_names[variable_names].cat.rename_categories({'AsbShng': 'AsbestosShingles',\n",
    "                                                            'AsphShn': \"AsphaltShingles\",\n",
    "                                                            \"BrkComm\": \"CommonBrick\",\n",
    "                                                            \"BrkFace\": \"FaceBrick\", \n",
    "                                                            \"CBlock\": \"CinderBlock\",\n",
    "                                                            \"CemntBd\": \"Cement\",\n",
    "                                                            \"HdBoard\": \"Hardboard\",\n",
    "                                                            \"ImStucc\": \"ImitationStucco\",\n",
    "                                                            \"MetalSd\": \"Metal\",\n",
    "                                                            \"VinylSd\": \"Vinyl\",\n",
    "                                                            \"Wd Sdng\": \"WoodSiding\",\n",
    "                                                            \"WdShing\": \"WoodShingles\"}, inplace = True)\n",
    "    \n",
    "    dataset_names.HeatingType.cat.rename_categories({\"GasA\": 'GasAir', \"GasW\": \"GasWater\", \"Grav\": \"Gravity\", \"OthW\": \"OtherWater\"}, inplace = True)\n",
    "    dataset_names.CentralAir.cat.rename_categories({\"N\": 'No', \"Y\": \"Yes\"}, inplace = True)\n",
    "    dataset_names.FoundationType.cat.rename_categories({\"BrkTil\": 'BrickTile', \"CBlock\": \"Cinder\", \"PConc\": \"Concrete\"}, inplace = True)\n",
    "    dataset_names.SaleFlags.cat.rename_categories({\"Abnorml\": 'Abnormal', \"AdjLand\": \"Adjoining\", \"Alloca\": \"Linked\", \"Partial\":\"Unfinished\"}, inplace = True)\n",
    "    dataset_names.FenceQual.cat.rename_categories({\"GdPriv\": 'GoodPrivacy', \"MnPriv\": \"MinimumPrivacy\", \"GdWo\": \"GoodWood\", \"MnWw\":\"MinimumWoodWire\"}, inplace = True)\n",
    "    dataset_names.MiscFeature.cat.rename_categories({\"Elev\": 'Elevator', \"Gar2\": \"2ndGarage\", \"Othr\": \"Other\", \"TenC\": \"Tennis\"}, inplace = True)\n",
    "    dataset_names.PavedDriveType.cat.rename_categories({\"Y\": 'Paved', \"P\": \"PartialPaved\", \"N\": \"NotPaved\"}, inplace = True)\n",
    "    dataset_names.StreetMaterial.cat.rename_categories({\"Grvl\": 'Gravel', \"Pave\": \"Paved\"}, inplace = True)\n",
    "    dataset_names.AlleyMaterial.cat.rename_categories({\"Grvl\": 'Gravel', \"Pave\": \"Paved\", \"NA\": \"NoAlley\"}, inplace = True)\n",
    "    dataset_names.Functional.cat.rename_categories({\"Typ\": 'Typical', \"Min1\": \"MinorDeduction1\", \"Min2\": \"MinorDeduction2\", \"Mod\": \"ModerateDeduction\", \"Maj1\": \"MajorDeductions1\", \"Maj2\": \"MajorDeduction2\", \"Sev\": \"SevereDamage\", \"Sal\": \"Salvage\"}, inplace = True)\n",
    "    dataset_names.RoofMaterial.cat.rename_categories({\"ClyTile\": 'ClayTile', \"CompShg\": \"Composite\", \"Membran\": \"Membrane\", \"Tar&Grv\":\"Tar&Gravel\", \"WdShake\": \"WoodShakes\", \"WdShngl\": \"WoodShingle\"}, inplace = True)\n",
    "    dataset_names.LandContour.cat.rename_categories({\"Lvl\": 'Level', \"Bnk\": \"Banked\", \"HLS\": \"Hillside\", \"Low\": \"Depression\"}, inplace = True)\n",
    "    dataset_names.LotConfig.cat.rename_categories({\"CulDSac\": 'CulDeSac', \"FR2\": \"Frontage2Sides\", \"FR3\": \"Frontage3Sides\"}, inplace = True)\n",
    "    dataset_names.PropertyType2.cat.rename_categories({\"1Fam\": '1Family', \"2FmCon\": \"2FamilyConv\", \"Duplx\": \"Duplex\", \"TwnhsE\": \"EndTownhouse\", \"TwnhsI\": \"InsideTownhouse\"}, inplace = True)\n",
    "    dataset_names.ZoningClass.cat.rename_categories({\"A\": 'Agriculture', \"C\": \"Commercial\", \"FV\": \"FloatingVillage\", \"I\": \"Industrial\", \"RH\": \"ResHighDens\", \"RL\": \"ResLowDens\", \"RP\": \"ResLowDensPark\", \"RM\": \"ResMedDens\"}, inplace = True)\n",
    "    dataset_names.VeneerType.cat.rename_categories({\"BrkCmn\": 'CommonBrick', \"BrkFace\": \"FaceBrick\", \"CBlock\": \"CinderBlock\"}, inplace = True)\n",
    "    dataset_names.ElectricalType.cat.rename_categories({\"SBrkr\": 'Good', \"FuseA\": \"Average\", \"FuseF\": \"Fair\", \"FuseP\": \"Poor\", \"Mix\": \"Mixture\"}, inplace = True)\n",
    "    dataset_names.HouseStyle.cat.rename_categories({\"1.5Fin\": '1.5Finished', \"1.5Unf\": \"1.5Unfinished\", \"2.5Fin\": \"2.5Finished\", \"2.5Unf\": \"2.5Unfinished\", \"SFoyer\": \"SplitFoyer\", \"SLvl\":\"SplitLevel\"}, inplace = True)\n",
    "    dataset_names.SaleType.cat.rename_categories({\"WD\": \"ConventionalWarranty\", \"CWD\": \"CashWarranty\", \"VWD\": \"VAWarranty\", \"COD\": \"CourtOfficer\", \"Con\": \"Contract\", \"ConLw\": \"ContractLowDownandInterest\", \"ConLI\": \"ContractLowInterest\", \"ConLD\": \"ContractLowDown\", \"Oth\": \"Other\"})\n",
    "    dataset_names.GarageType.cat.rename_categories({\"Attchd\": \"Attached\", \"Basment\": \"Basement\", \"Detchd\": \"Detached\"})    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Initial modelling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# valid_rmse = []\n",
    "\n",
    "# model = LinearRegression()\n",
    "\n",
    "\n",
    "# for i in range(3):\n",
    "#     trainX, validX, trainY, validY = ms.train_test_split(Xtrain_elastic, Ytrain_elastic, test_size=1.0/3, random_state=i)\n",
    "#     model.fit(trainX, trainY)\n",
    "#     new_Y = model.predict(validX)\n",
    "#     valid_rmse.append(rmse(new_Y, validY))\n",
    "\n",
    "# print('Training Error: {}'.format(np.round(np.array(train_error), 4))) # training error\n",
    "# print('Test Error: {}'.format(np.round(np.array(test_error), 4))) # test error\n",
    "# print('Training R Squared: {}'.format(np.round(np.array(train_rsq), 4)))\n",
    "# print('Test R Squared: {}'.format(np.round(np.array(test_rsq), 4)))\n",
    "\n",
    "# # calculating the mean of the test error\n",
    "# print('Test Error Mean: {:6f}'.format(np.mean(test_error)))\n",
    "\n",
    "# # calculating the standard deviation of the test errors\n",
    "# print('Test Error SD: {:6f}'.format(np.std(test_error)))\n",
    "\n",
    "# # calculating test r squared\n",
    "# print('Mean R Squared: {:6f}'.format(np.mean(test_rsq)))\n",
    "\n",
    "# y = lasso.predict(Xtrain_elastic)\n",
    "# ydf = pd.DataFrame(y)\n",
    "# rmse(ydf, pd.DataFrame(Ytrain_elastic))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((1060, 374), (1060,), (400, 374), (400,))"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create simple fixed validation set for consistent tests across models\n",
    "df, y, nas = proc_df(train, 'SalePrice', max_n_cat=100) # Uses one hot encoding up to 100 categories\n",
    "n_valid = 400  \n",
    "n_trn = len(df)-n_valid\n",
    "#df_raw2 = df_raw2.sample(frac=1) # Randomise order for simple validation test\n",
    "#df = df.sample(frac=1)\n",
    "\n",
    "X_train, X_valid = split_vals(df, n_trn)\n",
    "y_train, y_valid = split_vals(y, n_trn)\n",
    "\n",
    "X_train.shape, y_train.shape, X_valid.shape, y_valid.shape\n",
    "\n",
    "# [X_train, y_train, X_valid, y_valid, na_dict] = proc_df2(category_list = \"none\", df1 = train, df2 = None, y_fld = \"SalePrice\", n_validation = 400, percentage_validation = None, add_variables = False, imputation_methodology = \"median\", impute_using = \"all\", drop_outliers = False, log_size_variables = False, variables_to_square = None, drop_variables = False)pr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Decision tree:\n",
      "CPU times: user 112 ms, sys: 16 ms, total: 128 ms\n",
      "Wall time: 221 ms\n",
      "[0.21168554305506201, 0.2312282585974743, 0.7291719312935654, 0.6269187955728921]\n",
      "-----------\n",
      "RF result:\n",
      "CPU times: user 15.2 s, sys: 388 ms, total: 15.6 s\n",
      "Wall time: 4.9 s\n",
      "[0.07867746764368241, 0.13902779218875663, 0.9625879056821336, 0.8651270659571741, 0.8728805220729544]\n",
      "-----------\n",
      "Extra trees result:\n",
      "CPU times: user 12.2 s, sys: 252 ms, total: 12.5 s\n",
      "Wall time: 3.75 s\n",
      "[0.08671525931675882, 0.13872948364959695, 0.954553302580496, 0.8657052321354394]\n",
      "-----------\n",
      "GBM result:\n",
      "CPU times: user 804 ms, sys: 0 ns, total: 804 ms\n",
      "Wall time: 801 ms\n",
      "[0.042719979699722015, 0.12590413373982307, 0.9889700549259194, 0.8893881816520747]\n",
      "-----------\n",
      "XGBoost:\n",
      "[02:21:17] WARNING: /workspace/src/objective/regression_obj.cu:152: reg:linear is now deprecated in favor of reg:squarederror.\n",
      "CPU times: user 664 ms, sys: 8 ms, total: 672 ms\n",
      "Wall time: 679 ms\n",
      "[0.0977720732128695, 0.13412887815187818, 0.9422248710993286, 0.8744646211566299]\n"
     ]
    },
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.38.0 (20140413.2041)\n",
       " -->\n",
       "<!-- Title: Tree Pages: 1 -->\n",
       "<svg width=\"720pt\" height=\"434pt\"\n",
       " viewBox=\"0.00 0.00 720.00 434.49\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(0.778659 0.778659) rotate(0) translate(4 554)\">\n",
       "<title>Tree</title>\n",
       "<polygon fill=\"white\" stroke=\"none\" points=\"-4,4 -4,-554 920.667,-554 920.667,4 -4,4\"/>\n",
       "<!-- 0 -->\n",
       "<g id=\"node1\" class=\"node\"><title>0</title>\n",
       "<polygon fill=\"#f4caab\" stroke=\"black\" points=\"135.667,-308.5 19.6667,-308.5 19.6667,-240.5 135.667,-240.5 135.667,-308.5\"/>\n",
       "<text text-anchor=\"start\" x=\"27.6667\" y=\"-293.3\" font-family=\"Times,serif\" font-size=\"14.00\">OverallQual ≤ 6.5</text>\n",
       "<text text-anchor=\"start\" x=\"46.1667\" y=\"-278.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.17</text>\n",
       "<text text-anchor=\"start\" x=\"33.6667\" y=\"-263.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1060</text>\n",
       "<text text-anchor=\"start\" x=\"39.6667\" y=\"-248.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.03</text>\n",
       "</g>\n",
       "<!-- 1 -->\n",
       "<g id=\"node2\" class=\"node\"><title>1</title>\n",
       "<polygon fill=\"#f9e0ce\" stroke=\"black\" points=\"413.667,-361.5 233.667,-361.5 233.667,-293.5 413.667,-293.5 413.667,-361.5\"/>\n",
       "<text text-anchor=\"start\" x=\"241.667\" y=\"-346.3\" font-family=\"Times,serif\" font-size=\"14.00\">AboveGradeAreaSF ≤ 1378.5</text>\n",
       "<text text-anchor=\"start\" x=\"292.167\" y=\"-331.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.09</text>\n",
       "<text text-anchor=\"start\" x=\"283.167\" y=\"-316.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 653</text>\n",
       "<text text-anchor=\"start\" x=\"285.667\" y=\"-301.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.81</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;1 -->\n",
       "<g id=\"edge1\" class=\"edge\"><title>0&#45;&gt;1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M135.874,-286.926C161.943,-292.589 193.574,-299.459 223.338,-305.924\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"222.835,-309.397 233.35,-308.099 224.32,-302.556 222.835,-309.397\"/>\n",
       "<text text-anchor=\"middle\" x=\"212.322\" y=\"-317.922\" font-family=\"Times,serif\" font-size=\"14.00\">True</text>\n",
       "</g>\n",
       "<!-- 8 -->\n",
       "<g id=\"node9\" class=\"node\"><title>8</title>\n",
       "<polygon fill=\"#eda674\" stroke=\"black\" points=\"381.667,-255.5 265.667,-255.5 265.667,-187.5 381.667,-187.5 381.667,-255.5\"/>\n",
       "<text text-anchor=\"start\" x=\"273.667\" y=\"-240.3\" font-family=\"Times,serif\" font-size=\"14.00\">OverallQual ≤ 7.5</text>\n",
       "<text text-anchor=\"start\" x=\"292.167\" y=\"-225.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.09</text>\n",
       "<text text-anchor=\"start\" x=\"283.167\" y=\"-210.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 407</text>\n",
       "<text text-anchor=\"start\" x=\"285.667\" y=\"-195.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.38</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;8 -->\n",
       "<g id=\"edge8\" class=\"edge\"><title>0&#45;&gt;8</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M135.874,-262.074C171.691,-254.294 218.004,-244.234 255.553,-236.078\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"256.367,-239.483 265.396,-233.94 254.881,-232.642 256.367,-239.483\"/>\n",
       "<text text-anchor=\"middle\" x=\"244.369\" y=\"-216.717\" font-family=\"Times,serif\" font-size=\"14.00\">False</text>\n",
       "</g>\n",
       "<!-- 2 -->\n",
       "<g id=\"node3\" class=\"node\"><title>2</title>\n",
       "<polygon fill=\"#fbebdf\" stroke=\"black\" points=\"683.667,-486.5 547.667,-486.5 547.667,-418.5 683.667,-418.5 683.667,-486.5\"/>\n",
       "<text text-anchor=\"start\" x=\"555.667\" y=\"-471.3\" font-family=\"Times,serif\" font-size=\"14.00\">TotalBsmtSF ≤ 814.5</text>\n",
       "<text text-anchor=\"start\" x=\"584.167\" y=\"-456.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.07</text>\n",
       "<text text-anchor=\"start\" x=\"575.167\" y=\"-441.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 407</text>\n",
       "<text text-anchor=\"start\" x=\"580.667\" y=\"-426.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.7</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;2 -->\n",
       "<g id=\"edge2\" class=\"edge\"><title>1&#45;&gt;2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M403.91,-361.657C445.695,-379.667 496.721,-401.661 538.1,-419.497\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"536.746,-422.725 547.315,-423.469 539.517,-416.296 536.746,-422.725\"/>\n",
       "</g>\n",
       "<!-- 5 -->\n",
       "<g id=\"node6\" class=\"node\"><title>5</title>\n",
       "<polygon fill=\"#f5ceb2\" stroke=\"black\" points=\"708.167,-361.5 523.167,-361.5 523.167,-293.5 708.167,-293.5 708.167,-361.5\"/>\n",
       "<text text-anchor=\"start\" x=\"531.167\" y=\"-346.3\" font-family=\"Times,serif\" font-size=\"14.00\">FinishedBasementSF1 ≤ 239.0</text>\n",
       "<text text-anchor=\"start\" x=\"584.167\" y=\"-331.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.05</text>\n",
       "<text text-anchor=\"start\" x=\"575.167\" y=\"-316.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 246</text>\n",
       "<text text-anchor=\"start\" x=\"577.667\" y=\"-301.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.99</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;5 -->\n",
       "<g id=\"edge5\" class=\"edge\"><title>1&#45;&gt;5</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M414.043,-327.5C445.278,-327.5 480.573,-327.5 512.753,-327.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"513.083,-331 523.083,-327.5 513.083,-324 513.083,-331\"/>\n",
       "</g>\n",
       "<!-- 3 -->\n",
       "<g id=\"node4\" class=\"node\"><title>3</title>\n",
       "<polygon fill=\"#ffffff\" stroke=\"black\" points=\"900.167,-550 803.167,-550 803.167,-497 900.167,-497 900.167,-550\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-534.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.09</text>\n",
       "<text text-anchor=\"start\" x=\"811.167\" y=\"-519.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 139</text>\n",
       "<text text-anchor=\"start\" x=\"816.667\" y=\"-504.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.5</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;3 -->\n",
       "<g id=\"edge3\" class=\"edge\"><title>2&#45;&gt;3</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M683.722,-472.846C718.236,-483.318 759.889,-495.956 792.954,-505.989\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"792.338,-509.459 802.923,-509.014 794.37,-502.761 792.338,-509.459\"/>\n",
       "</g>\n",
       "<!-- 4 -->\n",
       "<g id=\"node5\" class=\"node\"><title>4</title>\n",
       "<polygon fill=\"#f9e1cf\" stroke=\"black\" points=\"900.167,-479 803.167,-479 803.167,-426 900.167,-426 900.167,-479\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-463.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.04</text>\n",
       "<text text-anchor=\"start\" x=\"811.167\" y=\"-448.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 268</text>\n",
       "<text text-anchor=\"start\" x=\"816.667\" y=\"-433.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.8</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;4 -->\n",
       "<g id=\"edge4\" class=\"edge\"><title>2&#45;&gt;4</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M683.722,-452.5C718.087,-452.5 759.53,-452.5 792.526,-452.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"792.923,-456 802.923,-452.5 792.923,-449 792.923,-456\"/>\n",
       "</g>\n",
       "<!-- 6 -->\n",
       "<g id=\"node7\" class=\"node\"><title>6</title>\n",
       "<polygon fill=\"#f8dbc6\" stroke=\"black\" points=\"900.167,-408 803.167,-408 803.167,-355 900.167,-355 900.167,-408\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-392.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.05</text>\n",
       "<text text-anchor=\"start\" x=\"811.167\" y=\"-377.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 104</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-362.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 11.86</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;6 -->\n",
       "<g id=\"edge6\" class=\"edge\"><title>5&#45;&gt;6</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M708.478,-348.687C736.963,-355.261 767.539,-362.317 793.011,-368.195\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"792.337,-371.631 802.868,-370.47 793.911,-364.81 792.337,-371.631\"/>\n",
       "</g>\n",
       "<!-- 7 -->\n",
       "<g id=\"node8\" class=\"node\"><title>7</title>\n",
       "<polygon fill=\"#f3c4a3\" stroke=\"black\" points=\"900.167,-337 803.167,-337 803.167,-284 900.167,-284 900.167,-337\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-321.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.03</text>\n",
       "<text text-anchor=\"start\" x=\"811.167\" y=\"-306.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 142</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-291.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.08</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;7 -->\n",
       "<g id=\"edge7\" class=\"edge\"><title>5&#45;&gt;7</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M708.478,-320.83C736.837,-318.77 767.268,-316.559 792.672,-314.713\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"793.148,-318.188 802.868,-313.973 792.641,-311.206 793.148,-318.188\"/>\n",
       "</g>\n",
       "<!-- 9 -->\n",
       "<g id=\"node10\" class=\"node\"><title>9</title>\n",
       "<polygon fill=\"#f0b68d\" stroke=\"black\" points=\"705.667,-255.5 525.667,-255.5 525.667,-187.5 705.667,-187.5 705.667,-255.5\"/>\n",
       "<text text-anchor=\"start\" x=\"533.667\" y=\"-240.3\" font-family=\"Times,serif\" font-size=\"14.00\">AboveGradeAreaSF ≤ 1725.0</text>\n",
       "<text text-anchor=\"start\" x=\"584.167\" y=\"-225.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.04</text>\n",
       "<text text-anchor=\"start\" x=\"575.167\" y=\"-210.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 236</text>\n",
       "<text text-anchor=\"start\" x=\"577.667\" y=\"-195.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.23</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;9 -->\n",
       "<g id=\"edge9\" class=\"edge\"><title>8&#45;&gt;9</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M381.961,-221.5C419.94,-221.5 470.777,-221.5 515.246,-221.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"515.513,-225 525.513,-221.5 515.513,-218 515.513,-225\"/>\n",
       "</g>\n",
       "<!-- 12 -->\n",
       "<g id=\"node13\" class=\"node\"><title>12</title>\n",
       "<polygon fill=\"#e89151\" stroke=\"black\" points=\"705.667,-131.5 525.667,-131.5 525.667,-63.5 705.667,-63.5 705.667,-131.5\"/>\n",
       "<text text-anchor=\"start\" x=\"533.667\" y=\"-116.3\" font-family=\"Times,serif\" font-size=\"14.00\">AboveGradeAreaSF ≤ 1939.5</text>\n",
       "<text text-anchor=\"start\" x=\"584.167\" y=\"-101.3\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.07</text>\n",
       "<text text-anchor=\"start\" x=\"575.167\" y=\"-86.3\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 171</text>\n",
       "<text text-anchor=\"start\" x=\"580.667\" y=\"-71.3\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.6</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;12 -->\n",
       "<g id=\"edge12\" class=\"edge\"><title>8&#45;&gt;12</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M381.961,-197.002C422.97,-179.467 478.972,-155.521 525.761,-135.515\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"527.17,-138.719 534.988,-131.569 524.418,-132.283 527.17,-138.719\"/>\n",
       "</g>\n",
       "<!-- 10 -->\n",
       "<g id=\"node11\" class=\"node\"><title>10</title>\n",
       "<polygon fill=\"#f2bf9b\" stroke=\"black\" points=\"900.167,-266 803.167,-266 803.167,-213 900.167,-213 900.167,-266\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-250.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.03</text>\n",
       "<text text-anchor=\"start\" x=\"811.167\" y=\"-235.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 138</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-220.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.13</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;10 -->\n",
       "<g id=\"edge10\" class=\"edge\"><title>9&#45;&gt;10</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M705.8,-228.356C734.989,-230.602 766.635,-233.036 792.883,-235.055\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"792.792,-238.558 803.031,-235.836 793.329,-231.579 792.792,-238.558\"/>\n",
       "</g>\n",
       "<!-- 11 -->\n",
       "<g id=\"node12\" class=\"node\"><title>11</title>\n",
       "<polygon fill=\"#eda978\" stroke=\"black\" points=\"897.667,-195 805.667,-195 805.667,-142 897.667,-142 897.667,-195\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-179.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.04</text>\n",
       "<text text-anchor=\"start\" x=\"814.667\" y=\"-164.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 98</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-149.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.35</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;11 -->\n",
       "<g id=\"edge11\" class=\"edge\"><title>9&#45;&gt;11</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M705.8,-201.312C736.034,-194.464 768.905,-187.019 795.681,-180.954\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"796.647,-184.324 805.627,-178.701 795.101,-177.497 796.647,-184.324\"/>\n",
       "</g>\n",
       "<!-- 13 -->\n",
       "<g id=\"node14\" class=\"node\"><title>13</title>\n",
       "<polygon fill=\"#eb9e67\" stroke=\"black\" points=\"897.667,-124 805.667,-124 805.667,-71 897.667,-71 897.667,-124\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-108.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.04</text>\n",
       "<text text-anchor=\"start\" x=\"814.667\" y=\"-93.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 90</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-78.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.46</text>\n",
       "</g>\n",
       "<!-- 12&#45;&gt;13 -->\n",
       "<g id=\"edge13\" class=\"edge\"><title>12&#45;&gt;13</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M705.8,-97.5C735.901,-97.5 768.616,-97.5 795.327,-97.5\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"795.627,-101 805.627,-97.5 795.627,-94.0001 795.627,-101\"/>\n",
       "</g>\n",
       "<!-- 14 -->\n",
       "<g id=\"node15\" class=\"node\"><title>14</title>\n",
       "<polygon fill=\"#e58139\" stroke=\"black\" points=\"897.667,-53 805.667,-53 805.667,-0 897.667,-0 897.667,-53\"/>\n",
       "<text text-anchor=\"start\" x=\"820.167\" y=\"-37.8\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.06</text>\n",
       "<text text-anchor=\"start\" x=\"814.667\" y=\"-22.8\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 81</text>\n",
       "<text text-anchor=\"start\" x=\"813.667\" y=\"-7.8\" font-family=\"Times,serif\" font-size=\"14.00\">value = 12.76</text>\n",
       "</g>\n",
       "<!-- 12&#45;&gt;14 -->\n",
       "<g id=\"edge14\" class=\"edge\"><title>12&#45;&gt;14</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M705.8,-70.4554C736.168,-61.2412 769.195,-51.22 796.035,-43.0763\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"797.074,-46.4186 805.627,-40.1659 795.042,-39.7202 797.074,-46.4186\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.files.Source at 0x7fda1c8fd588>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7fda17f6fb00>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create baseline models for single decison tree, RF, ExtraTrees, GBM\n",
    "\n",
    "dec_tree = RandomForestRegressor(n_estimators=1, max_depth=3, bootstrap=False, n_jobs=-1)\n",
    "print(\"Decision tree:\")\n",
    "%time dec_tree.fit(X_train, y_train)\n",
    "print_score(dec_tree)\n",
    "print(\"-----------\")\n",
    "\n",
    "rf = RandomForestRegressor(n_estimators=1000, n_jobs=-1,min_samples_leaf=3, max_features=0.5, oob_score=True)\n",
    "print(\"RF result:\")\n",
    "%time rf.fit(X_train, y_train)\n",
    "print_score(rf)\n",
    "print(\"-----------\")\n",
    "\n",
    "rf_extra = ExtraTreesRegressor(n_estimators=1000, n_jobs=-1,min_samples_leaf=5, max_features=0.5)\n",
    "print(\"Extra trees result:\")\n",
    "%time rf_extra.fit(X_train, y_train)\n",
    "print_score(rf_extra)\n",
    "print(\"-----------\")\n",
    "\n",
    "gbm =  GradientBoostingRegressor(n_estimators= 1000, learning_rate = 0.1, max_features ='sqrt', min_samples_leaf= 10) #, oob_score=True)\n",
    "print(\"GBM result:\")\n",
    "%time gbm.fit(X_train, y_train)\n",
    "print_score(gbm)\n",
    "print(\"-----------\")\n",
    "\n",
    "xgb = xgboost.XGBRegressor(colsample_bytree=0.4,\n",
    "                 gamma=0,                 \n",
    "                 learning_rate=0.1,\n",
    "                 max_depth=3,\n",
    "                 min_child_weight=1.5,\n",
    "                 n_estimators=100,                                                                    \n",
    "                 reg_alpha=0.75,\n",
    "                 reg_lambda=0.45,\n",
    "                 subsample=0.6,\n",
    "                 seed=1) \n",
    "print(\"XGBoost:\")\n",
    "%time xgb.fit(X_train, y_train)\n",
    "print_score(xgb)\n",
    "\n",
    "# Visualise tree\n",
    "draw_tree(dec_tree.estimators_[0], df, precision=2)\n",
    "\n",
    "# Visualise feature importace for random forest\n",
    "fi0 = rf_feat_importance(rf, df); fi0[:15]\n",
    "plot_fi(fi0[:40])\n",
    "#fi1 = rf_feat_importance(m1, df); fi[:10]\n",
    "\n",
    "# Visualise feature importace for GBM\n",
    "fi0 = rf_feat_importance(gbm, df); fi0[:15]\n",
    "plot_fi(fi0[:40])\n",
    "#fi1 = rf_feat_importance(m1, df); fi[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Feature correlations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "27\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x1080 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Simplify to focus on most important features with > 0.5% feature importance, invetsigate correlations\n",
    "\n",
    "to_keep = fi0[fi0.imp>0.01].cols\n",
    "print(len(to_keep))\n",
    "df_keep = df[to_keep].copy()\n",
    "\n",
    "corr = np.round(scipy.stats.spearmanr(df_keep).correlation, 4)\n",
    "corr_condensed = hc.distance.squareform(1-corr)\n",
    "z = hc.linkage(corr_condensed, method='average')\n",
    "fig = plt.figure(figsize=(20,15))\n",
    "dendrogram = hc.dendrogram(z, labels=df_keep.columns, orientation='left', leaf_font_size=16)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tree  hyperparameter tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_max_features</th>\n",
       "      <th>param_min_samples_leaf</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>6.458563</td>\n",
       "      <td>0.038291</td>\n",
       "      <td>0.149108</td>\n",
       "      <td>0.003635</td>\n",
       "      <td>0.2</td>\n",
       "      <td>2</td>\n",
       "      <td>{'max_features': 0.2, 'min_samples_leaf': 2}</td>\n",
       "      <td>-0.016082</td>\n",
       "      <td>-0.025501</td>\n",
       "      <td>-0.02467</td>\n",
       "      <td>-0.017729</td>\n",
       "      <td>-0.020437</td>\n",
       "      <td>-0.020884</td>\n",
       "      <td>0.003711</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.001528</td>\n",
       "      <td>-0.001358</td>\n",
       "      <td>-0.001352</td>\n",
       "      <td>-0.001508</td>\n",
       "      <td>-0.001527</td>\n",
       "      <td>-0.001455</td>\n",
       "      <td>0.000082</td>\n",
       "      <td>0.144512</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0       6.458563      0.038291         0.149108        0.003635   \n",
       "\n",
       "  param_max_features param_min_samples_leaf  \\\n",
       "0                0.2                      2   \n",
       "\n",
       "                                         params  split0_test_score  \\\n",
       "0  {'max_features': 0.2, 'min_samples_leaf': 2}          -0.016082   \n",
       "\n",
       "   split1_test_score  split2_test_score  split3_test_score  split4_test_score  \\\n",
       "0          -0.025501           -0.02467          -0.017729          -0.020437   \n",
       "\n",
       "   mean_test_score  std_test_score  rank_test_score  split0_train_score  \\\n",
       "0        -0.020884        0.003711                1           -0.001528   \n",
       "\n",
       "   split1_train_score  split2_train_score  split3_train_score  \\\n",
       "0           -0.001358           -0.001352           -0.001508   \n",
       "\n",
       "   split4_train_score  mean_train_score  std_train_score      RMSE  \n",
       "0           -0.001527         -0.001455         0.000082  0.144512  "
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Grid search for extra random forest with n trees\n",
    "erf =  ExtraTreesRegressor(n_estimators= 1000)\n",
    "\n",
    "# Feature space searched\n",
    "#grid_param_erf = [{'min_samples_leaf': [2,3,5,7,10,20], 'max_features' :  ['auto', 'sqrt', 'log2', 0.5, 0.2, 0.1]}]\n",
    "\n",
    "# Best params\n",
    "grid_param_erf = [{'min_samples_leaf': [2], 'max_features' :  [0.2]}]\n",
    "para_search_erf = ms.GridSearchCV(estimator=erf, param_grid=grid_param_erf, scoring='neg_mean_squared_error', cv=5, return_train_score=True)\n",
    "para_search_erf = para_search_erf.fit(df, y)\n",
    "tuning_results_erf = pd.DataFrame(para_search_erf.cv_results_)\n",
    "tuning_results_erf['RMSE'] = np.sqrt(abs(tuning_results_erf.mean_test_score))\n",
    "tuning_results_erf.sort_values('RMSE').head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_max_features</th>\n",
       "      <th>param_min_samples_leaf</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>6.412342</td>\n",
       "      <td>0.011614</td>\n",
       "      <td>0.131881</td>\n",
       "      <td>0.002189</td>\n",
       "      <td>0.2</td>\n",
       "      <td>2</td>\n",
       "      <td>{'max_features': 0.2, 'min_samples_leaf': 2}</td>\n",
       "      <td>-0.015916</td>\n",
       "      <td>-0.023328</td>\n",
       "      <td>-0.021121</td>\n",
       "      <td>-0.016389</td>\n",
       "      <td>-0.020433</td>\n",
       "      <td>-0.019437</td>\n",
       "      <td>0.002852</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.004747</td>\n",
       "      <td>-0.004432</td>\n",
       "      <td>-0.004482</td>\n",
       "      <td>-0.004867</td>\n",
       "      <td>-0.004679</td>\n",
       "      <td>-0.004641</td>\n",
       "      <td>0.000163</td>\n",
       "      <td>0.139418</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0       6.412342      0.011614         0.131881        0.002189   \n",
       "\n",
       "  param_max_features param_min_samples_leaf  \\\n",
       "0                0.2                      2   \n",
       "\n",
       "                                         params  split0_test_score  \\\n",
       "0  {'max_features': 0.2, 'min_samples_leaf': 2}          -0.015916   \n",
       "\n",
       "   split1_test_score  split2_test_score  split3_test_score  split4_test_score  \\\n",
       "0          -0.023328          -0.021121          -0.016389          -0.020433   \n",
       "\n",
       "   mean_test_score  std_test_score  rank_test_score  split0_train_score  \\\n",
       "0        -0.019437        0.002852                1           -0.004747   \n",
       "\n",
       "   split1_train_score  split2_train_score  split3_train_score  \\\n",
       "0           -0.004432           -0.004482           -0.004867   \n",
       "\n",
       "   split4_train_score  mean_train_score  std_train_score      RMSE  \n",
       "0           -0.004679         -0.004641         0.000163  0.139418  "
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Grid search for random forest with n trees\n",
    "rf =  RandomForestRegressor(n_estimators= 1000)\n",
    "\n",
    "# Feature space searched\n",
    "#grid_param_rf = [{'min_samples_leaf': [2,3,5,7,10,20], 'max_features' :  ['auto', 'sqrt', 'log2', 0.5, 0.2, 0.1]}]\n",
    "#{'max_features': 0.2, 'min_samples_leaf': 2}\t0.064613\t0.134249\n",
    "\n",
    "# Best params\n",
    "grid_param_rf = [{'min_samples_leaf': [2], 'max_features' :  [0.2]}]\n",
    "para_search_rf = ms.GridSearchCV(estimator=rf, param_grid=grid_param_rf, scoring='neg_mean_squared_error', cv=5, return_train_score=True)\n",
    "para_search_rf = para_search_rf.fit(df, y)\n",
    "tuning_results_rf = pd.DataFrame(para_search_rf.cv_results_)\n",
    "tuning_results_rf['RMSE'] = np.sqrt(abs(tuning_results_rf.mean_test_score))\n",
    "tuning_results_rf.sort_values('RMSE').head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_learning_rate</th>\n",
       "      <th>param_max_features</th>\n",
       "      <th>param_min_samples_leaf</th>\n",
       "      <th>param_n_estimators</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "      <th>av_RMSE</th>\n",
       "      <th>min_RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2.264377</td>\n",
       "      <td>0.078808</td>\n",
       "      <td>0.025329</td>\n",
       "      <td>0.000366</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.1</td>\n",
       "      <td>3</td>\n",
       "      <td>1750</td>\n",
       "      <td>{'learning_rate': 0.05, 'max_features': 0.1, '...</td>\n",
       "      <td>-0.011164</td>\n",
       "      <td>-0.01793</td>\n",
       "      <td>-0.015567</td>\n",
       "      <td>-0.010739</td>\n",
       "      <td>-0.015582</td>\n",
       "      <td>-0.014196</td>\n",
       "      <td>0.002789</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.000838</td>\n",
       "      <td>-0.000832</td>\n",
       "      <td>-0.000807</td>\n",
       "      <td>-0.000892</td>\n",
       "      <td>-0.000868</td>\n",
       "      <td>-0.000847</td>\n",
       "      <td>0.00003</td>\n",
       "      <td>0.119148</td>\n",
       "      <td>0.133904</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0       2.264377      0.078808         0.025329        0.000366   \n",
       "\n",
       "  param_learning_rate param_max_features param_min_samples_leaf  \\\n",
       "0                0.05                0.1                      3   \n",
       "\n",
       "  param_n_estimators                                             params  \\\n",
       "0               1750  {'learning_rate': 0.05, 'max_features': 0.1, '...   \n",
       "\n",
       "   split0_test_score  split1_test_score  split2_test_score  split3_test_score  \\\n",
       "0          -0.011164           -0.01793          -0.015567          -0.010739   \n",
       "\n",
       "   split4_test_score  mean_test_score  std_test_score  rank_test_score  \\\n",
       "0          -0.015582        -0.014196        0.002789                1   \n",
       "\n",
       "   split0_train_score  split1_train_score  split2_train_score  \\\n",
       "0           -0.000838           -0.000832           -0.000807   \n",
       "\n",
       "   split3_train_score  split4_train_score  mean_train_score  std_train_score  \\\n",
       "0           -0.000892           -0.000868         -0.000847          0.00003   \n",
       "\n",
       "    av_RMSE  min_RMSE  \n",
       "0  0.119148  0.133904  "
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Grid search for GBM\n",
    "# Best params {'min_samples_leaf': 3, 'max_features' : 0.1, 'n_estimators': 1200, learning_rate: 0.03]  -> 0.114849\n",
    "# Feature space searched\n",
    "#grid_param_gbm = [{'min_samples_leaf': [2,3,5,7,10,20], 'max_features' : ['auto', 'sqrt', 'log2', 0.5, 0.2, 0.1], 'n_estimators': [100,1000,10000]}] \n",
    "               #'learning_rate ' : [0.001, 0.01, 0.1]}]\n",
    "\n",
    "# Best params\n",
    "gbm =  GradientBoostingRegressor()\n",
    "grid_param_gbm = [{'min_samples_leaf': [3], 'max_features' : [0.1], 'n_estimators': [1750], 'learning_rate': [0.05]}]\n",
    "para_search_gbm = ms.GridSearchCV(estimator=gbm, param_grid=grid_param_gbm, scoring='neg_mean_squared_error', cv=5, return_train_score=True)\n",
    "para_search_gbm = para_search_gbm.fit(df, y)\n",
    "tuning_results_gbm = pd.DataFrame(para_search_gbm.cv_results_)\n",
    "tuning_results_gbm['av_RMSE'] = np.sqrt(abs(tuning_results_gbm.mean_test_score))\n",
    "tuning_results_gbm['min_RMSE'] = np.sqrt(-tuning_results_gbm[['split0_test_score','split1_test_score','split2_test_score','split3_test_score','split4_test_score']].min(axis=1))\n",
    "tuning_results_gbm.sort_values('min_RMSE').head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 0 ns, sys: 0 ns, total: 0 ns\n",
      "Wall time: 48.2 µs\n",
      "Fitting 3 folds for each of 1 candidates, totalling 3 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/sklearn/model_selection/_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   3 out of   3 | elapsed:   12.9s remaining:    0.0s\n",
      "[Parallel(n_jobs=-1)]: Done   3 out of   3 | elapsed:   12.9s finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[02:23:13] WARNING: /workspace/src/objective/regression_obj.cu:152: reg:linear is now deprecated in favor of reg:squarederror.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_colsample_bytree</th>\n",
       "      <th>param_gamma</th>\n",
       "      <th>param_max_depth</th>\n",
       "      <th>param_min_child_weight</th>\n",
       "      <th>param_n_estimators</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>9.794068</td>\n",
       "      <td>0.259421</td>\n",
       "      <td>0.050839</td>\n",
       "      <td>0.002972</td>\n",
       "      <td>0.8</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>1000</td>\n",
       "      <td>{'colsample_bytree': 0.8, 'gamma': 0, 'max_dep...</td>\n",
       "      <td>-0.013286</td>\n",
       "      <td>-0.018812</td>\n",
       "      <td>-0.014527</td>\n",
       "      <td>-0.015542</td>\n",
       "      <td>0.002367</td>\n",
       "      <td>1</td>\n",
       "      <td>0.124667</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0       9.794068      0.259421         0.050839        0.002972   \n",
       "\n",
       "  param_colsample_bytree param_gamma param_max_depth param_min_child_weight  \\\n",
       "0                    0.8           0               3                      1   \n",
       "\n",
       "  param_n_estimators                                             params  \\\n",
       "0               1000  {'colsample_bytree': 0.8, 'gamma': 0, 'max_dep...   \n",
       "\n",
       "   split0_test_score  split1_test_score  split2_test_score  mean_test_score  \\\n",
       "0          -0.013286          -0.018812          -0.014527        -0.015542   \n",
       "\n",
       "   std_test_score  rank_test_score      RMSE  \n",
       "0        0.002367                1  0.124667  "
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Best params so far: colsample of 0.8,  min_child_weight =1, max_depth = 3, n_estimators = 1000, gamme = 0 -> \n",
    "\n",
    "# Feature space explored \n",
    "#    'colsample_bytree':[0.4,0.6,0.8], # Control overfitting\n",
    "#    'min_child_weight':[1,3,6,10], # Control overfitting\n",
    "#    'n_estimators': [100,1000,10000],\n",
    "#    'max_depth':[3,5,7,10],\n",
    "#    'gamma':[0,1,10]  # Higher values control model complexity and overfitting\n",
    "\n",
    "param_grid_xgb = {\n",
    "    'colsample_bytree':[0.8], # Control overfitting\n",
    "    'min_child_weight':[1], # Control overfitting\n",
    "    'n_estimators': [1000],\n",
    "    'max_depth':[3],\n",
    "    'gamma':[0]  # Higher values control model complexity and overfitting\n",
    "}\n",
    "                    \n",
    "%time xgb_model = xgboost.XGBRegressor(seed=1)\n",
    "para_search_xgb = ms.GridSearchCV(estimator = xgb_model, param_grid = param_grid_xgb, n_jobs=-1,iid=False, verbose=10,scoring='neg_mean_squared_error')\n",
    "para_search_xgb.fit(df, y)\n",
    "tuning_results_xgb = pd.DataFrame(para_search_xgb.cv_results_)\n",
    "tuning_results_xgb['RMSE'] = np.sqrt(abs(tuning_results_xgb.mean_test_score))\n",
    "tuning_results_xgb.sort_values('RMSE')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Ridge vs Lasso"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "#elastic_train = train.copy() # Making copy\n",
    "#Xtrain_elastic, Ytrain_elastic, na_elastic = proc_df(elastic_train, 'SalePrice', max_n_cat=200) #Splitting with saleprice\n",
    "#### [trainX, trainY, validX, validY, na_scale] = proc_df2(category_list = \"none\", df1 = train, df2 = None, y_fld = \"SalePrice\", n_validation = 300, percentage_validation = None, add_variables = False, imputation_methodology = \"median\", impute_using = \"all\", drop_outliers = False, log_size_variables = False, variables_to_square = None, drop_variables = False)\n",
    "\n",
    "#y_df = pd.DataFrame(Ytrain_elastic)\n",
    "\n",
    "#### Arbitrary removal of 300 observations for validation\n",
    "#np.random.seed(1)\n",
    "#validIdxes = np.random.choice(range(1460), size= 300, replace=False)\n",
    "#trainIdxes = list(set(range(1460))-set(validIdxes))\n",
    "#trainX = Xtrain_elastic.iloc[trainIdxes]\n",
    "#validX  = Xtrain_elastic.iloc[validIdxes]\n",
    "#trainY = y_df.iloc[trainIdxes]\n",
    "#validY  = y_df.iloc[validIdxes]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "ridge = Ridge()\n",
    "lasso = Lasso()\n",
    "net   = ElasticNet()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Ridge\n",
    "\n",
    "#ridge.set_params(normalize=True)\n",
    "#ridge_scores_train = []\n",
    "#ridge_scores_valid  = []\n",
    "\n",
    "#alphas = np.linspace(1,11,50)\n",
    "\n",
    "#for alpha in alphas:\n",
    "#            ridge.set_params(alpha=alpha)\n",
    "#            ridge.fit(trainX, trainY)\n",
    "#            ridge_scores_train.append(ridge.score(trainX, trainY))\n",
    "#            ridge_scores_valid.append(ridge.score(validX, validY))\n",
    "#ridge_scores_train = np.array(ridge_scores_train) \n",
    "#ridge_scores_valid = np.array(ridge_scores_valid)\n",
    "\n",
    "#### Ridge, training vs validation set\n",
    "#plt.plot(alphas, ridge_scores_train, label=r'$train\\ R^2$')\n",
    "#plt.plot(alphas, ridge_scores_valid, label=r'$valid\\ R^2$')\n",
    "#plt.legend(loc=1)\n",
    "#plt.title(r'Ridge Train-Validation $R^2$ Comparison')\n",
    "#ridge_underfit = ridge_scores_valid < ridge_scores_train\n",
    "#last_underfit  = np.max(alphas[ridge_underfit])\n",
    "#plt.axvline(last_underfit, linestyle='--', color='g', label='optimal lambda', alpha=0.4)\n",
    "#plt.legend(loc=1)\n",
    "#plt.xlabel(r'hyperparameter $\\lambda$')\n",
    "#plt.ylabel(r'$R^2$')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Lasso\n",
    "\n",
    "#lasso.set_params(normalize=True)\n",
    "#lasso_scores_train = []\n",
    "#lasso_scores_valid  = []\n",
    "\n",
    "#alphas = np.linspace(0.00001, 0.0005, 50)\n",
    "\n",
    "#for alpha in alphas:\n",
    "#            lasso.set_params(alpha=alpha)\n",
    "#            lasso.fit(trainX, trainY)\n",
    "#            lasso_scores_train.append(lasso.score(trainX, trainY))\n",
    "#            lasso_scores_valid.append(lasso.score(validX, validY))\n",
    "#lasso_scores_train = np.array(lasso_scores_train) \n",
    "#lasso_scores_valid = np.array(lasso_scores_valid)\n",
    "\n",
    "#### Lasso training vs validation set \n",
    "#plt.plot(alphas, lasso_scores_train, label=r'$train\\ R^2$')\n",
    "#plt.plot(alphas, lasso_scores_valid, label=r'$valid\\ R^2$')\n",
    "#plt.title(r'Lasso Train-Validation $R^2$ Comparison')\n",
    "#lasso_underfit = lasso_scores_valid < lasso_scores_train\n",
    "#last_underfit  = np.max(alphas[lasso_underfit])\n",
    "#plt.axvline(last_underfit, linestyle='--', color='g', label='optimal lambda', alpha=0.4)\n",
    "#plt.legend(loc=1)\n",
    "#plt.xlabel(r'hyperparameter $\\lambda$')\n",
    "#plt.ylabel(r'$R^2$')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Cross Validation, Lasso"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Finding optimum number of folds\n",
    "#mean_error = []\n",
    "#std_error = []\n",
    "#kfold = range(3, 15)\n",
    "#for i in kfold:\n",
    "#    scores = ms.cross_val_score(estimator=lasso, X=Xtrain_elastic, y=Ytrain_elastic, cv=i)\n",
    "#    mean_error.append(1 - scores.mean())\n",
    "#    std_error.append(scores.std())\n",
    "\n",
    "#s1 = plt.scatter(kfold, mean_error, label='mean error')\n",
    "#plt.plot(kfold, mean_error)\n",
    "#s2 = plt.scatter(kfold, std_error, color='red', label='std')\n",
    "#plt.plot(kfold, std_error, color='red')\n",
    "#plt.xlim(0, 15)\n",
    "#plt.xlabel('number of fold')\n",
    "#plt.legend(handles=[s1, s2], loc=2)\n",
    "#plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Continue with 3 folds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import sklearn.model_selection as ms\n",
    "#trainX, validX, trainY, validY = ms.train_test_split(Xtrain_elastic, Ytrain_elastic, \n",
    "#                                                       test_size=1.0/5, random_state=0)\n",
    "#print('Original: {}, {}'.format(Xtrain_elastic.shape, Ytrain_elastic.shape))\n",
    "#print('Training: {}, {}'.format(trainX.shape, trainY.shape))\n",
    "#print('Test: {}, {}'.format(validX.shape, validY.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Finding best lambda with 3-fold CV\n",
    "#lasso.set_params(normalize=True, max_iter=10000)\n",
    "#alphas = np.linspace(0.00001, 0.0005, 50)\n",
    "#lasso_scores_train = []\n",
    "#lasso_scores_valid  = []\n",
    "#score_means = []\n",
    "\n",
    "#for alpha in alphas:\n",
    "#    lasso.set_params(alpha = alpha)\n",
    "#    lasso.fit(Xtrain_elastic, Ytrain_elastic)\n",
    "#    scores = ms.cross_val_score(estimator=lasso, X=Xtrain_elastic, y=Ytrain_elastic, cv=3)\n",
    "#    score_means.append(scores.mean())\n",
    "    \n",
    "#plt.plot(alphas, score_means)\n",
    "#plt.title('Best Alpha with 5-fold CV')\n",
    "#plt.ylabel('R^2')\n",
    "#plt.xlabel('Alphas')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Best Lambda of 0.0001\n",
    "#lasso.set_params(alpha=0.0001, normalize=True)\n",
    "\n",
    "#train_error = []\n",
    "#test_error = []\n",
    "#train_rsq = []\n",
    "#test_rsq = []\n",
    "\n",
    "#for i in range(5):\n",
    "#    trainX, validX, trainY, validY = ms.train_test_split(Xtrain_elastic, Ytrain_elastic, test_size=1.0/3, random_state=i)\n",
    "#    lasso.fit(trainX, trainY)\n",
    "#    train_error.append(1 - lasso.score(trainX, trainY))\n",
    "#    test_error.append(1 - lasso.score(validX, validY))\n",
    "#    train_rsq.append(lasso.score(trainX, trainY))\n",
    "#    test_rsq.append(lasso.score(validX, validY))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "#print('Training Error: {}'.format(np.round(np.array(train_error), 4))) # training error\n",
    "#print('Test Error: {}'.format(np.round(np.array(test_error), 4))) # test error\n",
    "#print('Training R Squared: {}'.format(np.round(np.array(train_rsq), 4)))\n",
    "#print('Test R Squared: {}'.format(np.round(np.array(test_rsq), 4)))\n",
    "\n",
    "## calculating the mean of the test error\n",
    "#print('Test Error Mean: {:6f}'.format(np.mean(test_error)))\n",
    "\n",
    "## calculating the standard deviation of the test errors\n",
    "#print('Test Error SD: {:6f}'.format(np.std(test_error)))\n",
    "\n",
    "## calculating test r squared\n",
    "#print('Mean R Squared: {:6f}'.format(np.mean(test_rsq)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# RMSE with Lasso and lambda 0.0001\n",
    "#y = lasso.predict(Xtrain_elastic)\n",
    "#ydf = pd.DataFrame(y)\n",
    "#rmse(ydf, pd.DataFrame(Ytrain_elastic))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Finding lasso coefficients with alpha 0.0001\n",
    "#lasso.set_params(alpha=0.0001, normalize=True)\n",
    "#lasso.fit(Xtrain_elastic, Ytrain_elastic)\n",
    "\n",
    "#coefs_lasso  = []\n",
    "\n",
    "#coefs_lasso = lasso.coef_\n",
    "#coefs_lasso = pd.DataFrame(coefs_lasso, index = Xtrain_elastic.columns)\n",
    "#coefs_lasso.columns = ['Coefficient']\n",
    "#coefs_lasso = coefs_lasso.sort_values(by = 'Coefficient', ascending=False)\n",
    "#coefs_lasso"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "#coefs_lasso.index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Variables that lasso removes\n",
    "#len(coefs_lasso[coefs_lasso['Coefficient']==0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Variables lasso keeps\n",
    "#len(coefs_lasso[coefs_lasso['Coefficient']!=0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Elastic Net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Elastic net with cv 3, lambda 0.00035 - 0.0004, rho 0.85 - 0.95\n",
    "#N_alpha = 2\n",
    "#N_rho   = 11\n",
    "#alphaRange = np.linspace(0.00035, 0.0004, N_alpha)\n",
    "#rhoRange   = np.linspace(0.85,0.95, N_rho)\n",
    "#scores     = np.zeros((N_rho, N_alpha))\n",
    "#for alphaIdx, alpha in enumerate(alphaRange):\n",
    "#    for rhoIdx, rho in enumerate(rhoRange):\n",
    "#            net.set_params(alpha=alpha, l1_ratio=rho, max_iter=1000)\n",
    "#            net.fit(Xtrain_elastic, Ytrain_elastic)\n",
    "#            score = ms.cross_val_score(estimator=net, X=Xtrain_elastic, y=Ytrain_elastic, cv=3).mean()\n",
    "#            scores[rhoIdx, alphaIdx] = score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "#rhoRange"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "#alphaRange"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "#scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "#grid_df = pd.DataFrame(scores, columns = [alphaRange], index = [rhoRange])\n",
    "#grid_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Move forward with alpha as 0.0004 and rho as 0.91"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "#net.set_params(alpha=0.0004, l1_ratio=0.91, normalize=False, max_iter=30000)\n",
    "#net.fit(Xtrain_elastic, Ytrain_elastic)\n",
    "#cv_scores = ms.cross_val_score(estimator=net, X=Xtrain_elastic, y=Ytrain_elastic, cv=3)\n",
    "#cv_scores.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "# RMSE with Net and lambda 0.0004, Rho 0.91, cv 3\n",
    "#y = net.predict(Xtrain_elastic)\n",
    "#ydf = pd.DataFrame(y, )\n",
    "#rmse(ydf, pd.DataFrame(Ytrain_elastic))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "#zdf = pd.DataFrame(Ytrain_elastic)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "#mean_squared_error = mse(ydf, zdf)\n",
    "#mean_squared_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "#squared_error = se(ydf, zdf)\n",
    "#squared_error.columns = ['Squared Error']\n",
    "#squared_error['Above Average'] = squared_error['Squared Error'].apply(lambda x: 1 if x > 0.010669 else 0) \n",
    "#squared_error.columns = ['Squared Error', 'Above Average']\n",
    "#squared_error.index = squared_error.index+1\n",
    "# Biggest offenders on Squared Error with elastic net model\n",
    "#squared_error.sort_values(by='Squared Error', ascending=False).head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "#plt.hist(squared_error[squared_error['Squared Error']<0.1]['Squared Error'], bins=100)\n",
    "#plt.title('Distribution of squared error with elastic net model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Proportion of data set which has a squared error greater than the average.\n",
    "#len(squared_error[squared_error['Squared Error']>0.011356])/len(squared_error['Squared Error'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "# add column to training set with 'above average' which matches on index\n",
    "#error_search_df = Xtrain_elastic.copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "#new_df = error_search_df.merge(squared_error, how='outer', left_index=True, right_index=True)\n",
    "#new_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "#plt.scatter(x = new_df.YearBuilt, y=new_df['Squared Error'], c=new_df['Above Average'], alpha = 0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "#plt.scatter(x = new_df.LotArea, y=new_df['Squared Error'], c=new_df['Above Average'], alpha = 0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "#plt.scatter(x = new_df.LotFrontage, y=new_df['Squared Error'], c=new_df['Above Average'], alpha = 0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "#len(squared_error[squared_error['Squared Error']>0.05])/len(squared_error['Squared Error'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Looking for clustering within variables where the top 2.8 percentile of Squared Error\n",
    "#import seaborn as sns\n",
    "#column_list = Xtrain_elastic.columns.to_list()\n",
    "#pp = sns.pairplot(data=new_df[new_df['Squared Error']>0.05],\n",
    "#                  x_vars=['Squared Error'],\n",
    "#                  y_vars=column_list[0:40])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Lasso with Proc DF2**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dropping 5 SD Outliers...\n",
      "Comprehensively dropping variables.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'Alphas')"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# One version of Processed Data Frame with following characteristics:\n",
    "\n",
    "[x, y, nas] = proc_df2(vars_to_cat=\"none\",\n",
    "                        df1 = train,\n",
    "                        df2 = None,\n",
    "                        y_fld = \"SalePrice\",\n",
    "                        n_validation=None,\n",
    "                        add_variables=True,\n",
    "                        imputation_methodology=\"mean\",\n",
    "                        impute_using=\"train\",\n",
    "                        drop_outliers=\"5SD\",\n",
    "                        log_size_variables=True,\n",
    "                        square_variables = True,\n",
    "                        drop_variables=\"comprehensive\")                                             \n",
    "\n",
    "\n",
    "# Finding best lambda with 3-fold CV\n",
    "\n",
    "lasso.set_params(normalize=True, max_iter=10000)\n",
    "alphas = np.linspace(0.00001, 0.0005, 50)\n",
    "lasso_scores_train = []\n",
    "lasso_scores_valid  = []\n",
    "score_means = []\n",
    "\n",
    "for alpha in alphas:\n",
    "    lasso.set_params(alpha = alpha)\n",
    "    lasso.fit(x, y)\n",
    "    scores = ms.cross_val_score(estimator=lasso, X=x, y=y, cv=3)\n",
    "    score_means.append(scores.mean())\n",
    "    \n",
    "plt.plot(alphas, score_means)\n",
    "plt.title('Best Alpha with 3-fold CV')\n",
    "plt.ylabel('R^2')\n",
    "plt.xlabel('Alphas')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dropping 5 SD Outliers...\n",
      "Comprehensively dropping variables.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_alpha</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>0.126389</td>\n",
       "      <td>0.101685</td>\n",
       "      <td>0.002153</td>\n",
       "      <td>0.000031</td>\n",
       "      <td>0.000136207</td>\n",
       "      <td>{'alpha': 0.00013620689655172413}</td>\n",
       "      <td>-0.011650</td>\n",
       "      <td>-0.014401</td>\n",
       "      <td>-0.011812</td>\n",
       "      <td>-0.012621</td>\n",
       "      <td>0.001260</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.010529</td>\n",
       "      <td>-0.009674</td>\n",
       "      <td>-0.010663</td>\n",
       "      <td>-0.010289</td>\n",
       "      <td>0.000438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>0.132390</td>\n",
       "      <td>0.109087</td>\n",
       "      <td>0.002107</td>\n",
       "      <td>0.000086</td>\n",
       "      <td>0.000132759</td>\n",
       "      <td>{'alpha': 0.00013275862068965518}</td>\n",
       "      <td>-0.011661</td>\n",
       "      <td>-0.014397</td>\n",
       "      <td>-0.011810</td>\n",
       "      <td>-0.012623</td>\n",
       "      <td>0.001256</td>\n",
       "      <td>2</td>\n",
       "      <td>-0.010487</td>\n",
       "      <td>-0.009637</td>\n",
       "      <td>-0.010622</td>\n",
       "      <td>-0.010249</td>\n",
       "      <td>0.000436</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>0.119091</td>\n",
       "      <td>0.096172</td>\n",
       "      <td>0.002017</td>\n",
       "      <td>0.000137</td>\n",
       "      <td>0.000139655</td>\n",
       "      <td>{'alpha': 0.0001396551724137931}</td>\n",
       "      <td>-0.011653</td>\n",
       "      <td>-0.014406</td>\n",
       "      <td>-0.011814</td>\n",
       "      <td>-0.012624</td>\n",
       "      <td>0.001261</td>\n",
       "      <td>3</td>\n",
       "      <td>-0.010569</td>\n",
       "      <td>-0.009711</td>\n",
       "      <td>-0.010703</td>\n",
       "      <td>-0.010328</td>\n",
       "      <td>0.000440</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>0.117391</td>\n",
       "      <td>0.095791</td>\n",
       "      <td>0.002148</td>\n",
       "      <td>0.000265</td>\n",
       "      <td>0.000143103</td>\n",
       "      <td>{'alpha': 0.00014310344827586207}</td>\n",
       "      <td>-0.011656</td>\n",
       "      <td>-0.014412</td>\n",
       "      <td>-0.011818</td>\n",
       "      <td>-0.012629</td>\n",
       "      <td>0.001263</td>\n",
       "      <td>4</td>\n",
       "      <td>-0.010609</td>\n",
       "      <td>-0.009748</td>\n",
       "      <td>-0.010743</td>\n",
       "      <td>-0.010367</td>\n",
       "      <td>0.000441</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>0.134534</td>\n",
       "      <td>0.111112</td>\n",
       "      <td>0.002113</td>\n",
       "      <td>0.000207</td>\n",
       "      <td>0.00012931</td>\n",
       "      <td>{'alpha': 0.0001293103448275862}</td>\n",
       "      <td>-0.011683</td>\n",
       "      <td>-0.014395</td>\n",
       "      <td>-0.011811</td>\n",
       "      <td>-0.012629</td>\n",
       "      <td>0.001249</td>\n",
       "      <td>5</td>\n",
       "      <td>-0.010442</td>\n",
       "      <td>-0.009600</td>\n",
       "      <td>-0.010580</td>\n",
       "      <td>-0.010208</td>\n",
       "      <td>0.000433</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    mean_fit_time  std_fit_time  mean_score_time  std_score_time  param_alpha  \\\n",
       "25       0.126389      0.101685         0.002153        0.000031  0.000136207   \n",
       "24       0.132390      0.109087         0.002107        0.000086  0.000132759   \n",
       "26       0.119091      0.096172         0.002017        0.000137  0.000139655   \n",
       "27       0.117391      0.095791         0.002148        0.000265  0.000143103   \n",
       "23       0.134534      0.111112         0.002113        0.000207   0.00012931   \n",
       "\n",
       "                               params  split0_test_score  split1_test_score  \\\n",
       "25  {'alpha': 0.00013620689655172413}          -0.011650          -0.014401   \n",
       "24  {'alpha': 0.00013275862068965518}          -0.011661          -0.014397   \n",
       "26   {'alpha': 0.0001396551724137931}          -0.011653          -0.014406   \n",
       "27  {'alpha': 0.00014310344827586207}          -0.011656          -0.014412   \n",
       "23   {'alpha': 0.0001293103448275862}          -0.011683          -0.014395   \n",
       "\n",
       "    split2_test_score  mean_test_score  std_test_score  rank_test_score  \\\n",
       "25          -0.011812        -0.012621        0.001260                1   \n",
       "24          -0.011810        -0.012623        0.001256                2   \n",
       "26          -0.011814        -0.012624        0.001261                3   \n",
       "27          -0.011818        -0.012629        0.001263                4   \n",
       "23          -0.011811        -0.012629        0.001249                5   \n",
       "\n",
       "    split0_train_score  split1_train_score  split2_train_score  \\\n",
       "25           -0.010529           -0.009674           -0.010663   \n",
       "24           -0.010487           -0.009637           -0.010622   \n",
       "26           -0.010569           -0.009711           -0.010703   \n",
       "27           -0.010609           -0.009748           -0.010743   \n",
       "23           -0.010442           -0.009600           -0.010580   \n",
       "\n",
       "    mean_train_score  std_train_score  \n",
       "25         -0.010289         0.000438  \n",
       "24         -0.010249         0.000436  \n",
       "26         -0.010328         0.000440  \n",
       "27         -0.010367         0.000441  \n",
       "23         -0.010208         0.000433  "
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Search for best alpha with a grid search\n",
    "\n",
    "[x, y, nas] = proc_df2(vars_to_cat=\"none\",\n",
    "            df1 = train,\n",
    "            df2 = None,\n",
    "            y_fld = \"SalePrice\",\n",
    "            n_validation=None,\n",
    "            add_variables=True,\n",
    "            imputation_methodology=\"mean\",\n",
    "            impute_using=\"train\",\n",
    "            drop_outliers=\"5SD\",\n",
    "            log_size_variables=True,\n",
    "            square_variables = True,\n",
    "            drop_variables=\"comprehensive\")\n",
    "\n",
    "# Grid search for best alpha\n",
    "\n",
    "lasso.set_params(normalize=True)\n",
    "lasso.fit(x, y)\n",
    "\n",
    "grid_param_lasso = [{'alpha': np.linspace(0.00005, 0.00015, 30)}]\n",
    "\n",
    "para_search_lasso = ms.GridSearchCV(estimator=lasso, param_grid=grid_param_lasso, scoring='neg_mean_squared_error', cv=3, return_train_score=True)\n",
    "#para_search_lasso = ms.GridSearchCV(estimator=lasso, param_grid=grid_param_lasso, scoring='r2', cv=3, return_train_score=True)\n",
    "para_search_lasso = para_search_lasso.fit(x, y)\n",
    "tuning_results_lasso = pd.DataFrame(para_search_lasso.cv_results_)\n",
    "tuning_results_lasso.sort_values('rank_test_score').head()\n",
    "\n",
    "# 0.000135 best alpha\n",
    "# 0.92 mean_r^2\n",
    "# 0.11 root mean sq error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dropping 5 SD Outliers...\n",
      "Comprehensively dropping variables.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Coefficient</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Residential</th>\n",
       "      <td>0.250435</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LotArea_logged</th>\n",
       "      <td>0.063895</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>OverallQual</th>\n",
       "      <td>0.053402</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Neigh1</th>\n",
       "      <td>0.051229</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Average_Score</th>\n",
       "      <td>0.040312</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Depression_Price_Flags</th>\n",
       "      <td>-0.052421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Moderate_Defects</th>\n",
       "      <td>-0.060236</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Major_Defects</th>\n",
       "      <td>-0.091558</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Exterior_CommonBrickProportion</th>\n",
       "      <td>-0.144259</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Worst_Defects</th>\n",
       "      <td>-0.208386</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>198 rows × 1 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                Coefficient\n",
       "Residential                        0.250435\n",
       "LotArea_logged                     0.063895\n",
       "OverallQual                        0.053402\n",
       "Neigh1                             0.051229\n",
       "Average_Score                      0.040312\n",
       "...                                     ...\n",
       "Depression_Price_Flags            -0.052421\n",
       "Moderate_Defects                  -0.060236\n",
       "Major_Defects                     -0.091558\n",
       "Exterior_CommonBrickProportion    -0.144259\n",
       "Worst_Defects                     -0.208386\n",
       "\n",
       "[198 rows x 1 columns]"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Lasso coefficients for this df:\n",
    "\n",
    "[x, y, nas] = proc_df2(vars_to_cat=\"none\",\n",
    "                        df1 = train,\n",
    "                        df2 = None,\n",
    "                        y_fld = \"SalePrice\",\n",
    "                        n_validation=None,\n",
    "                        add_variables=True,\n",
    "                        imputation_methodology=\"mean\",\n",
    "                        impute_using=\"train\",\n",
    "                        drop_outliers=\"5SD\",\n",
    "                        log_size_variables=True,\n",
    "                        square_variables = True,\n",
    "                        drop_variables=\"comprehensive\")\n",
    "\n",
    "# set best alpha with this df\n",
    "lasso.set_params(alpha=0.000135714, normalize=True)\n",
    "lasso.fit(x, y)\n",
    "\n",
    "coefs_lasso  = []\n",
    "\n",
    "coefs_lasso = lasso.coef_\n",
    "coefs_lasso = pd.DataFrame(coefs_lasso, index = x.columns)\n",
    "coefs_lasso.columns = ['Coefficient']\n",
    "coefs_lasso = coefs_lasso.sort_values(by = 'Coefficient', ascending=False)\n",
    "coefs_lasso\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.2536231884057971"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Proportion of variables it keeps:\n",
    "len(coefs_lasso[coefs_lasso['Coefficient']>=0])/(len(coefs_lasso[coefs_lasso['Coefficient']==0] + len(coefs_lasso[coefs_lasso['Coefficient']>=0])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model stacking - fitting residuals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dropping 5 SD Outliers...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:152: FutureWarning: Sorting because non-concatenation axis is not aligned. A future version\n",
      "of pandas will change to not sort by default.\n",
      "\n",
      "To accept the future behavior, pass 'sort=False'.\n",
      "\n",
      "To retain the current behavior and silence the warning, pass 'sort=True'.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Not dropping variables.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:152: FutureWarning: Sorting because non-concatenation axis is not aligned. A future version\n",
      "of pandas will change to not sort by default.\n",
      "\n",
      "To accept the future behavior, pass 'sort=False'.\n",
      "\n",
      "To retain the current behavior and silence the warning, pass 'sort=True'.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Minimally dropping variables.\n"
     ]
    }
   ],
   "source": [
    "#RF on RESIDUALS\n",
    "\n",
    "# Encode lasso data\n",
    "[x_train_lasso, y_train_lasso, x_test_lasso, y_test_lasso, nas] = proc_df2(vars_to_cat=\"none\",df1 = train,df2 = test,y_fld = \"SalePrice\",\n",
    "                                                                            n_validation=None,add_variables=True,imputation_methodology=\"median\",impute_using=\"train\",\n",
    "                                                                            drop_outliers=\"5SD\",log_size_variables=True,square_variables = False,drop_variables=False)\n",
    "\n",
    "lasso.set_params(alpha=0.00015, normalize=True)\n",
    "lasso.fit(x_train_lasso, y_train_lasso)\n",
    "\n",
    "lasso_results = pd.DataFrame(x_train_lasso.index, columns = ['Id'])\n",
    "lasso_results['lasso_pred'] = lasso.predict(x_train_lasso)\n",
    "lasso_results['actuals'] = y_train_lasso\n",
    "lasso_results['residuals'] = lasso_results.actuals - lasso_results.lasso_pred\n",
    "lasso_results.set_index('Id',inplace=True)\n",
    "\n",
    "# Encode RF data, need to drop outliers with ordinal/minimal features\n",
    "[x_train_rf, y_train_rf, x_test_rf, y_test_rf, nas_rf] = proc_df2(vars_to_cat=\"ordinal\",df1 = train,df2 = test,y_fld = \"SalePrice\",\n",
    "                                                    n_validation=None,add_variables=True,imputation_methodology=\"mean\",impute_using=\"train\",\n",
    "                                                    drop_outliers=False,log_size_variables=False,square_variables = False,drop_variables=\"minimal\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set up datasets for residual training, allowing for lasso prediction to be a feature in the estimation\n",
    "y_train_rf = np.array(lasso_results[\"residuals\"]).ravel()\n",
    "x_train_rf = pd.merge(lasso_results,x_train_rf, left_index = True, right_index = True, how='left').drop(['actuals','residuals'],axis=1)\n",
    "X_train = x_train_rf\n",
    "y_train = np.array(y_train_rf).ravel()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_max_features</th>\n",
       "      <th>param_min_samples_leaf</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "      <th>RMSE</th>\n",
       "      <th>min_RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.307179</td>\n",
       "      <td>0.017008</td>\n",
       "      <td>0.023135</td>\n",
       "      <td>0.001687</td>\n",
       "      <td>0.2</td>\n",
       "      <td>7</td>\n",
       "      <td>{'max_features': 0.2, 'min_samples_leaf': 7}</td>\n",
       "      <td>-0.009129</td>\n",
       "      <td>-0.010148</td>\n",
       "      <td>-0.011743</td>\n",
       "      <td>-0.008726</td>\n",
       "      <td>-0.009536</td>\n",
       "      <td>-0.009856</td>\n",
       "      <td>0.001053</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.00587</td>\n",
       "      <td>-0.005674</td>\n",
       "      <td>-0.005396</td>\n",
       "      <td>-0.005925</td>\n",
       "      <td>-0.005738</td>\n",
       "      <td>-0.005721</td>\n",
       "      <td>0.000185</td>\n",
       "      <td>0.099278</td>\n",
       "      <td>0.108366</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0       1.307179      0.017008         0.023135        0.001687   \n",
       "\n",
       "  param_max_features param_min_samples_leaf  \\\n",
       "0                0.2                      7   \n",
       "\n",
       "                                         params  split0_test_score  \\\n",
       "0  {'max_features': 0.2, 'min_samples_leaf': 7}          -0.009129   \n",
       "\n",
       "   split1_test_score  split2_test_score  split3_test_score  split4_test_score  \\\n",
       "0          -0.010148          -0.011743          -0.008726          -0.009536   \n",
       "\n",
       "   mean_test_score  std_test_score  rank_test_score  split0_train_score  \\\n",
       "0        -0.009856        0.001053                1            -0.00587   \n",
       "\n",
       "   split1_train_score  split2_train_score  split3_train_score  \\\n",
       "0           -0.005674           -0.005396           -0.005925   \n",
       "\n",
       "   split4_train_score  mean_train_score  std_train_score      RMSE  min_RMSE  \n",
       "0           -0.005738         -0.005721         0.000185  0.099278  0.108366  "
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Grid search for extra random forest with 1000 trees\n",
    "rf =  RandomForestRegressor(n_estimators= 200)\n",
    "\n",
    "# Feature space searched\n",
    "# grid_param_rf = [{'min_samples_leaf': [2,3,5,7,10,15,20], 'max_features' :  [0.2,0.1, 0.075, 'sqrt', 0.02]}]\n",
    "\n",
    "# Best params\n",
    "grid_param_rf = [{'min_samples_leaf': [7], 'max_features' :  [0.2]}]\n",
    "para_search_rf = ms.GridSearchCV(estimator=rf, param_grid=grid_param_rf, scoring='neg_mean_squared_error', cv=5, return_train_score=True)\n",
    "para_search_rf = para_search_rf.fit(X_train, y_train)\n",
    "tuning_results_rf = pd.DataFrame(para_search_rf.cv_results_)\n",
    "tuning_results_rf['RMSE'] = np.sqrt(abs(tuning_results_rf.mean_test_score))\n",
    "tuning_results_rf['min_RMSE'] = np.sqrt(-tuning_results_rf[['split0_test_score','split1_test_score','split2_test_score','split3_test_score','split4_test_score']].min(axis=1))\n",
    "tuning_results_rf.sort_values('RMSE').head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>mean_fit_time</th>\n",
       "      <th>std_fit_time</th>\n",
       "      <th>mean_score_time</th>\n",
       "      <th>std_score_time</th>\n",
       "      <th>param_max_features</th>\n",
       "      <th>param_min_samples_leaf</th>\n",
       "      <th>params</th>\n",
       "      <th>split0_test_score</th>\n",
       "      <th>split1_test_score</th>\n",
       "      <th>split2_test_score</th>\n",
       "      <th>split3_test_score</th>\n",
       "      <th>split4_test_score</th>\n",
       "      <th>mean_test_score</th>\n",
       "      <th>std_test_score</th>\n",
       "      <th>rank_test_score</th>\n",
       "      <th>split0_train_score</th>\n",
       "      <th>split1_train_score</th>\n",
       "      <th>split2_train_score</th>\n",
       "      <th>split3_train_score</th>\n",
       "      <th>split4_train_score</th>\n",
       "      <th>mean_train_score</th>\n",
       "      <th>std_train_score</th>\n",
       "      <th>RMSE</th>\n",
       "      <th>min_RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2.49058</td>\n",
       "      <td>0.076748</td>\n",
       "      <td>0.057175</td>\n",
       "      <td>0.000896</td>\n",
       "      <td>0.2</td>\n",
       "      <td>7</td>\n",
       "      <td>{'max_features': 0.2, 'min_samples_leaf': 7}</td>\n",
       "      <td>-0.009455</td>\n",
       "      <td>-0.010267</td>\n",
       "      <td>-0.012078</td>\n",
       "      <td>-0.008884</td>\n",
       "      <td>-0.009359</td>\n",
       "      <td>-0.010008</td>\n",
       "      <td>0.001126</td>\n",
       "      <td>1</td>\n",
       "      <td>-0.005393</td>\n",
       "      <td>-0.005211</td>\n",
       "      <td>-0.004921</td>\n",
       "      <td>-0.005443</td>\n",
       "      <td>-0.005265</td>\n",
       "      <td>-0.005247</td>\n",
       "      <td>0.000183</td>\n",
       "      <td>0.100041</td>\n",
       "      <td>0.109899</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   mean_fit_time  std_fit_time  mean_score_time  std_score_time  \\\n",
       "0        2.49058      0.076748         0.057175        0.000896   \n",
       "\n",
       "  param_max_features param_min_samples_leaf  \\\n",
       "0                0.2                      7   \n",
       "\n",
       "                                         params  split0_test_score  \\\n",
       "0  {'max_features': 0.2, 'min_samples_leaf': 7}          -0.009455   \n",
       "\n",
       "   split1_test_score  split2_test_score  split3_test_score  split4_test_score  \\\n",
       "0          -0.010267          -0.012078          -0.008884          -0.009359   \n",
       "\n",
       "   mean_test_score  std_test_score  rank_test_score  split0_train_score  \\\n",
       "0        -0.010008        0.001126                1           -0.005393   \n",
       "\n",
       "   split1_train_score  split2_train_score  split3_train_score  \\\n",
       "0           -0.005211           -0.004921           -0.005443   \n",
       "\n",
       "   split4_train_score  mean_train_score  std_train_score      RMSE  min_RMSE  \n",
       "0           -0.005265         -0.005247         0.000183  0.100041  0.109899  "
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Grid search for extra random forest with 1000 trees\n",
    "#grid_param_erf = [{'min_samples_leaf': [2,3,5,7,10,15,20], 'max_features' :  [0.2,0.1, 0.075, 'sqrt', 0.02]}]\n",
    "\n",
    "# Best params\n",
    "grid_param_erf = [{'min_samples_leaf': [7], 'max_features' :  [0.2]}]\n",
    "erf = ExtraTreesRegressor(n_estimators= 500)\n",
    "para_search_erf = ms.GridSearchCV(estimator=erf, param_grid=grid_param_erf, scoring='neg_mean_squared_error', cv=5, return_train_score=True)\n",
    "para_search_erf = para_search_erf.fit(X_train, y_train)\n",
    "tuning_results_erf = pd.DataFrame(para_search_erf.cv_results_)\n",
    "tuning_results_erf['RMSE'] = np.sqrt(abs(tuning_results_erf.mean_test_score))\n",
    "tuning_results_erf['min_RMSE'] = np.sqrt(-tuning_results_erf[['split0_test_score','split1_test_score','split2_test_score','split3_test_score','split4_test_score']].min(axis=1))\n",
    "tuning_results_erf.sort_values('RMSE').head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 24 ms, sys: 0 ns, total: 24 ms\n",
      "Wall time: 108 ms\n"
     ]
    },
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.38.0 (20140413.2041)\n",
       " -->\n",
       "<!-- Title: Tree Pages: 1 -->\n",
       "<svg width=\"720pt\" height=\"435pt\"\n",
       " viewBox=\"0.00 0.00 720.00 434.93\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(0.914867 0.914867) rotate(0) translate(4 471.4)\">\n",
       "<title>Tree</title>\n",
       "<polygon fill=\"white\" stroke=\"none\" points=\"-4,4 -4,-471.4 783,-471.4 783,4 -4,4\"/>\n",
       "<!-- 0 -->\n",
       "<g id=\"node1\" class=\"node\"><title>0</title>\n",
       "<polygon fill=\"#eca36e\" stroke=\"black\" points=\"128,-212.115 0,-212.115 0,-144.115 128,-144.115 128,-212.115\"/>\n",
       "<text text-anchor=\"start\" x=\"8\" y=\"-196.915\" font-family=\"Times,serif\" font-size=\"14.00\">lasso_pred ≤ 11.224</text>\n",
       "<text text-anchor=\"start\" x=\"32.5\" y=\"-181.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.01</text>\n",
       "<text text-anchor=\"start\" x=\"20\" y=\"-166.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1452</text>\n",
       "<text text-anchor=\"start\" x=\"32.5\" y=\"-151.915\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.0</text>\n",
       "</g>\n",
       "<!-- 1 -->\n",
       "<g id=\"node2\" class=\"node\"><title>1</title>\n",
       "<polygon fill=\"#f2be99\" stroke=\"black\" points=\"346.5,-272.115 227.5,-272.115 227.5,-204.115 346.5,-204.115 346.5,-272.115\"/>\n",
       "<text text-anchor=\"start\" x=\"235.5\" y=\"-256.915\" font-family=\"Times,serif\" font-size=\"14.00\">KitchenQual ≤ 3.5</text>\n",
       "<text text-anchor=\"start\" x=\"252\" y=\"-241.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.056</text>\n",
       "<text text-anchor=\"start\" x=\"250\" y=\"-226.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 19</text>\n",
       "<text text-anchor=\"start\" x=\"246.5\" y=\"-211.915\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.139</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;1 -->\n",
       "<g id=\"edge1\" class=\"edge\"><title>0&#45;&gt;1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M128.034,-195.228C155.995,-202.82 188.962,-211.77 217.526,-219.525\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"216.863,-222.971 227.43,-222.214 218.697,-216.216 216.863,-222.971\"/>\n",
       "<text text-anchor=\"middle\" x=\"205.739\" y=\"-230.942\" font-family=\"Times,serif\" font-size=\"14.00\">True</text>\n",
       "</g>\n",
       "<!-- 8 -->\n",
       "<g id=\"node9\" class=\"node\"><title>8</title>\n",
       "<polygon fill=\"#eca26d\" stroke=\"black\" points=\"410.5,-152.115 163.5,-152.115 163.5,-84.1151 410.5,-84.1151 410.5,-152.115\"/>\n",
       "<text text-anchor=\"start\" x=\"171.5\" y=\"-136.915\" font-family=\"Times,serif\" font-size=\"14.00\">Exterior_CommonBrickProportion ≤ 0.25</text>\n",
       "<text text-anchor=\"start\" x=\"252\" y=\"-121.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.009</text>\n",
       "<text text-anchor=\"start\" x=\"243\" y=\"-106.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1433</text>\n",
       "<text text-anchor=\"start\" x=\"249\" y=\"-91.9151\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.002</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;8 -->\n",
       "<g id=\"edge8\" class=\"edge\"><title>0&#45;&gt;8</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M128.034,-161.002C136.098,-158.812 144.579,-156.51 153.251,-154.156\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"154.377,-157.477 163.111,-151.479 152.543,-150.721 154.377,-157.477\"/>\n",
       "<text text-anchor=\"middle\" x=\"141.419\" y=\"-135.35\" font-family=\"Times,serif\" font-size=\"14.00\">False</text>\n",
       "</g>\n",
       "<!-- 2 -->\n",
       "<g id=\"node3\" class=\"node\"><title>2</title>\n",
       "<polygon fill=\"#f7d9c3\" stroke=\"black\" points=\"639,-392.115 447,-392.115 447,-324.115 639,-324.115 639,-392.115\"/>\n",
       "<text text-anchor=\"start\" x=\"455\" y=\"-376.915\" font-family=\"Times,serif\" font-size=\"14.00\">UnfinishedBasementSF ≤ 655.0</text>\n",
       "<text text-anchor=\"start\" x=\"511.5\" y=\"-361.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.04</text>\n",
       "<text text-anchor=\"start\" x=\"506\" y=\"-346.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 10</text>\n",
       "<text text-anchor=\"start\" x=\"502.5\" y=\"-331.915\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.276</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;2 -->\n",
       "<g id=\"edge2\" class=\"edge\"><title>1&#45;&gt;2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M346.592,-265.796C380.357,-281.749 423.319,-302.045 460.355,-319.543\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"459.098,-322.82 469.635,-323.927 462.088,-316.49 459.098,-322.82\"/>\n",
       "</g>\n",
       "<!-- 5 -->\n",
       "<g id=\"node6\" class=\"node\"><title>5</title>\n",
       "<polygon fill=\"#eba06a\" stroke=\"black\" points=\"598,-272.115 488,-272.115 488,-204.115 598,-204.115 598,-272.115\"/>\n",
       "<text text-anchor=\"start\" x=\"496\" y=\"-256.915\" font-family=\"Times,serif\" font-size=\"14.00\">1stFlrSF ≤ 746.5</text>\n",
       "<text text-anchor=\"start\" x=\"511.5\" y=\"-241.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.03</text>\n",
       "<text text-anchor=\"start\" x=\"509.5\" y=\"-226.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 9</text>\n",
       "<text text-anchor=\"start\" x=\"505\" y=\"-211.915\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.013</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;5 -->\n",
       "<g id=\"edge5\" class=\"edge\"><title>1&#45;&gt;5</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M346.592,-238.115C385.677,-238.115 437.083,-238.115 477.36,-238.115\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"477.609,-241.615 487.609,-238.115 477.609,-234.615 477.609,-241.615\"/>\n",
       "</g>\n",
       "<!-- 3 -->\n",
       "<g id=\"node4\" class=\"node\"><title>3</title>\n",
       "<polygon fill=\"#fef8f4\" stroke=\"black\" points=\"775.5,-464.615 678.5,-464.615 678.5,-411.615 775.5,-411.615 775.5,-464.615\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-449.415\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.022</text>\n",
       "<text text-anchor=\"start\" x=\"693.5\" y=\"-434.415\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 5</text>\n",
       "<text text-anchor=\"start\" x=\"686.5\" y=\"-419.415\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.431</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;3 -->\n",
       "<g id=\"edge3\" class=\"edge\"><title>2&#45;&gt;3</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M621.448,-392.158C637.472,-399.202 654.07,-406.498 669.193,-413.145\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"667.838,-416.372 678.401,-417.192 670.655,-409.964 667.838,-416.372\"/>\n",
       "</g>\n",
       "<!-- 4 -->\n",
       "<g id=\"node5\" class=\"node\"><title>4</title>\n",
       "<polygon fill=\"#f1ba93\" stroke=\"black\" points=\"772,-384.615 682,-384.615 682,-331.615 772,-331.615 772,-384.615\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-369.415\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.009</text>\n",
       "<text text-anchor=\"start\" x=\"693.5\" y=\"-354.415\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 5</text>\n",
       "<text text-anchor=\"start\" x=\"690\" y=\"-339.415\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.12</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;4 -->\n",
       "<g id=\"edge4\" class=\"edge\"><title>2&#45;&gt;4</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M639.264,-358.115C650.243,-358.115 661.14,-358.115 671.312,-358.115\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"671.623,-361.615 681.623,-358.115 671.623,-354.615 671.623,-361.615\"/>\n",
       "</g>\n",
       "<!-- 6 -->\n",
       "<g id=\"node7\" class=\"node\"><title>6</title>\n",
       "<polygon fill=\"#f1b991\" stroke=\"black\" points=\"775.5,-304.615 678.5,-304.615 678.5,-251.615 775.5,-251.615 775.5,-304.615\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-289.415\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.008</text>\n",
       "<text text-anchor=\"start\" x=\"693.5\" y=\"-274.415\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 5</text>\n",
       "<text text-anchor=\"start\" x=\"686.5\" y=\"-259.415\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.113</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;6 -->\n",
       "<g id=\"edge6\" class=\"edge\"><title>5&#45;&gt;6</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M598.113,-250.008C620.194,-254.861 645.762,-260.48 668.207,-265.413\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"667.757,-268.898 678.275,-267.626 669.26,-262.061 667.757,-268.898\"/>\n",
       "</g>\n",
       "<!-- 7 -->\n",
       "<g id=\"node8\" class=\"node\"><title>7</title>\n",
       "<polygon fill=\"#e58139\" stroke=\"black\" points=\"770,-224.615 684,-224.615 684,-171.615 770,-171.615 770,-224.615\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-209.415\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.013</text>\n",
       "<text text-anchor=\"start\" x=\"693.5\" y=\"-194.415\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 4</text>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-179.415\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.17</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;7 -->\n",
       "<g id=\"edge7\" class=\"edge\"><title>5&#45;&gt;7</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M598.113,-226.222C622.135,-220.943 650.284,-214.756 674.049,-209.533\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"674.925,-212.924 683.94,-207.359 673.422,-206.087 674.925,-212.924\"/>\n",
       "</g>\n",
       "<!-- 9 -->\n",
       "<g id=\"node10\" class=\"node\"><title>9</title>\n",
       "<polygon fill=\"#eca26d\" stroke=\"black\" points=\"613,-152.115 473,-152.115 473,-84.1151 613,-84.1151 613,-152.115\"/>\n",
       "<text text-anchor=\"start\" x=\"481\" y=\"-136.915\" font-family=\"Times,serif\" font-size=\"14.00\">GarageYrBlt ≤ 2007.5</text>\n",
       "<text text-anchor=\"start\" x=\"508\" y=\"-121.915\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.009</text>\n",
       "<text text-anchor=\"start\" x=\"499\" y=\"-106.915\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1432</text>\n",
       "<text text-anchor=\"start\" x=\"505\" y=\"-91.9151\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.002</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;9 -->\n",
       "<g id=\"edge9\" class=\"edge\"><title>8&#45;&gt;9</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M410.536,-118.115C428.29,-118.115 446.161,-118.115 462.759,-118.115\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"462.916,-121.615 472.916,-118.115 462.916,-114.615 462.916,-121.615\"/>\n",
       "</g>\n",
       "<!-- 12 -->\n",
       "<g id=\"node13\" class=\"node\"><title>12</title>\n",
       "<polygon fill=\"#ffffff\" stroke=\"black\" points=\"591.5,-56.6151 494.5,-56.6151 494.5,-3.61514 591.5,-3.61514 591.5,-56.6151\"/>\n",
       "<text text-anchor=\"start\" x=\"512.5\" y=\"-41.4151\" font-family=\"Times,serif\" font-size=\"14.00\">mse = &#45;0.0</text>\n",
       "<text text-anchor=\"start\" x=\"509.5\" y=\"-26.4151\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1</text>\n",
       "<text text-anchor=\"start\" x=\"502.5\" y=\"-11.4151\" font-family=\"Times,serif\" font-size=\"14.00\">value = &#45;0.468</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;12 -->\n",
       "<g id=\"edge12\" class=\"edge\"><title>8&#45;&gt;12</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M386.203,-84.092C419.565,-72.5334 455.727,-60.0052 484.805,-49.9307\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"486.077,-53.1941 494.381,-46.6132 483.786,-46.5798 486.077,-53.1941\"/>\n",
       "</g>\n",
       "<!-- 10 -->\n",
       "<g id=\"node11\" class=\"node\"><title>10</title>\n",
       "<polygon fill=\"#eca36e\" stroke=\"black\" points=\"779,-144.615 675,-144.615 675,-91.6151 779,-91.6151 779,-144.615\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-129.415\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.009</text>\n",
       "<text text-anchor=\"start\" x=\"683\" y=\"-114.415\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 1380</text>\n",
       "<text text-anchor=\"start\" x=\"695.5\" y=\"-99.4151\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.0</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;10 -->\n",
       "<g id=\"edge10\" class=\"edge\"><title>9&#45;&gt;10</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M613.066,-118.115C630.054,-118.115 648.127,-118.115 664.719,-118.115\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"664.836,-121.615 674.836,-118.115 664.836,-114.615 664.836,-121.615\"/>\n",
       "</g>\n",
       "<!-- 11 -->\n",
       "<g id=\"node12\" class=\"node\"><title>11</title>\n",
       "<polygon fill=\"#ea975c\" stroke=\"black\" points=\"773,-65.6151 681,-65.6151 681,-12.6151 773,-12.6151 773,-65.6151\"/>\n",
       "<text text-anchor=\"start\" x=\"692\" y=\"-50.4151\" font-family=\"Times,serif\" font-size=\"14.00\">mse = 0.008</text>\n",
       "<text text-anchor=\"start\" x=\"690\" y=\"-35.4151\" font-family=\"Times,serif\" font-size=\"14.00\">samples = 52</text>\n",
       "<text text-anchor=\"start\" x=\"689\" y=\"-20.4151\" font-family=\"Times,serif\" font-size=\"14.00\">value = 0.058</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;11 -->\n",
       "<g id=\"edge11\" class=\"edge\"><title>9&#45;&gt;11</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M613.066,-88.136C632.419,-79.7355 653.179,-70.7241 671.555,-62.748\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"673.147,-65.8725 680.926,-58.6801 670.36,-59.4513 673.147,-65.8725\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.files.Source at 0x7fda1786d1d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Simple decision tree visualisation\n",
    "\n",
    "dec_tree_resi = RandomForestRegressor(n_estimators=1, max_depth=3, bootstrap=False, n_jobs=-1)\n",
    "%time dec_tree_resi.fit(X_train, y_train)\n",
    "draw_tree(dec_tree_resi.estimators_[0], X_train, precision=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7fda178960b8>"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualise feature importace for random forest\n",
    "\n",
    "final_rf = RandomForestRegressor(n_estimators=10000, n_jobs=-1,min_samples_leaf=7, max_features=0.2, oob_score=True)\n",
    "final_rf.fit(X_train, y_train)\n",
    "#print_score(erf)\n",
    "final_rf.feature_importances_\n",
    "fi2 = rf_feat_importance(final_rf, X_train);\n",
    "plot_fi(fi2[:10])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 0.1091641299437824\n",
      "0.5 0.09515547113705859\n",
      "1.0 0.08224998490770981\n",
      "1.5 0.0710513635083222\n",
      "2.0 0.062484177233403014\n",
      "2.5 0.057732040166178375\n",
      "3.0 0.057744648479608036\n",
      "3.5 0.06251911950266169\n",
      "4.0 0.07110257441058679\n",
      "4.5 0.08231191761408736\n",
      "5.0 0.0952243003853214\n",
      "0.0 0.11148256898283679\n",
      "0.5 0.09724448339319536\n",
      "1.0 0.08410085110734569\n",
      "1.5 0.07264815212188436\n",
      "2.0 0.0638035430022737\n",
      "2.5 0.05875685011754409\n",
      "3.0 0.05849940183456689\n",
      "3.5 0.06308985526566639\n",
      "4.0 0.07160179399328877\n",
      "4.5 0.08283502406498124\n",
      "5.0 0.09583737665341575\n"
     ]
    }
   ],
   "source": [
    "# Define manual cross vaildation to optimise dampening parameter across i folds\n",
    "\n",
    "i=0\n",
    "while i<2: #Three fold validation\n",
    "    rand_shared = round(1000*random())\n",
    "    X_tr_lasso, X_val_lasso, y_tr_lasso, y_val_lasso = ms.train_test_split(x_train_lasso, y_train_lasso, test_size=0.2, random_state=rand_shared)\n",
    "   # tr_ids = pd.DataFrame(X_tr_lasso.index, columns = ['Id']).set_index('Id')\n",
    "   \n",
    "    lasso_tr = pd.DataFrame(X_tr_lasso.index, columns = ['Id'])\n",
    "    lasso_tr['lasso_pred'] = lasso.predict(X_tr_lasso)\n",
    "    lasso_tr.set_index('Id',inplace=True)\n",
    "    X_train_rf = pd.merge(lasso_tr,x_train_rf, left_index = True, right_index = True)\n",
    "    \n",
    "    lasso_val = pd.DataFrame(X_val_lasso.index, columns = ['Id'])\n",
    "#    lasso_val['lasso_pred'] = lasso.predict(X_val_lasso)\n",
    "    lasso_val.set_index('Id',inplace=True)\n",
    "    X_val_rf = pd.merge(lasso_val,x_train_rf, left_index = True, right_index = True)\n",
    "  \n",
    "\n",
    "    for adj_weight in np.linspace(0, 5, 11):\n",
    "        comb_predict = lasso.predict(X_val_lasso) + adj_weight * final_rf.predict(X_val_rf)\n",
    "        print(adj_weight, math.sqrt(((comb_predict-y_val_lasso)**2).mean()))\n",
    "    i +=1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Final Kaggle submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train actuals:                    0\n",
      "count    1452.000000\n",
      "mean   180376.891185\n",
      "std     79068.087698\n",
      "min     34900.000000\n",
      "25%    129900.000000\n",
      "50%    162900.000000\n",
      "75%    213492.500000\n",
      "max    755000.000000\n",
      "Test predictions: count    1.459000e+03\n",
      "mean     1.791700e+05\n",
      "std      8.433679e+04\n",
      "min      4.126351e+04\n",
      "25%      1.275183e+05\n",
      "50%      1.564828e+05\n",
      "75%      2.102604e+05\n",
      "max      1.425244e+06\n",
      "Name: SalePrice, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "# Residual weighting set to 1 \n",
    "adj_weight = 1\n",
    "\n",
    "test_predictions = pd.DataFrame(x_test_lasso.index, columns = ['Id'])\n",
    "test_predictions['lasso_pred'] = lasso.predict(x_test_lasso)\n",
    "test_predictions.set_index('Id',inplace=True)\n",
    "x_test_rf = pd.merge(test_predictions,x_test_rf, left_index = True, right_index = True, how='left')\n",
    "test_predictions[\"RF_adjustments\"] = final_rf.predict(x_test_rf) #RF adjustments\n",
    "test_predictions[\"final_estimate\"] = test_predictions[\"lasso_pred\"] + adj_weight * test_predictions[\"RF_adjustments\"]\n",
    "test_predictions[\"SalePrice\"] = np.exp(test_predictions[\"final_estimate\"].astype(\"float\"))\n",
    "test_predictions.reset_index(inplace=True)\n",
    "\n",
    "# Check output distibrution looks sensible\n",
    "print(\"Train actuals:\", pd.DataFrame(np.exp(y_train_lasso)).describe())\n",
    "print(\"Test predictions:\",test_predictions.SalePrice.describe())\n",
    "\n",
    "# Final submission \n",
    "test_submission = test_predictions[['Id','SalePrice']]\n",
    "test_submission.to_csv('submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}