examples/plot.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2'\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "import os\n",
    "import numpy as np\n",
    "import json\n",
    "import matplotlib.pyplot as plt\n",
    "import pickle\n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def merge_and_fill_trajectories(pandas_data_frames, default_value=None):\n",
    "    # merge all trajectories keeping all time steps\n",
    "    df = pd.DataFrame().join(pandas_data_frames, how='outer')\n",
    "\n",
    "    # forward fill to make it a propper step function\n",
    "    df = df.fillna(method='ffill')\n",
    "    if default_value is None:\n",
    "        # backward fill to replace the NaNs for the early times by the\n",
    "        # performance of a random configuration\n",
    "        df = df.fillna(method='bfill')\n",
    "    else:\n",
    "        df = df.fillna(default_value)\n",
    "    return df\n",
    "\n",
    "\n",
    "def get_trajectories(losses, iterations):\n",
    "    dfs = []\n",
    "    for i in range(len(losses)):\n",
    "        loss = losses[i]\n",
    "        iteration = iterations[i]\n",
    "        # print('Run %d, Min: %f'%(i, loss))\n",
    "        df = pd.DataFrame({str(i): loss}, index=iteration)\n",
    "        dfs.append(df)\n",
    "\n",
    "    df = merge_and_fill_trajectories(dfs, default_value=None)\n",
    "    if df.empty:\n",
    "        pass\n",
    "\n",
    "    return np.array(df.T), np.array(df.index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_results(predictor, path, epochs, metric='valid_acc', dataset='cifar10', ug=False):\n",
    "    \n",
    "    output = []\n",
    "    time = []\n",
    "    nan_count = 0\n",
    "    missing_results_count = 0\n",
    "    \n",
    "    algo_path = os.path.join(path, predictor)\n",
    "    for seed_dir in os.listdir(algo_path):\n",
    "        result_file = os.path.join(algo_path, seed_dir, 'errors.json')\n",
    "        result = json.load(open(result_file))\n",
    "        \n",
    "        config = result[0]\n",
    "        val_acc = result[1]['valid_acc'][:epochs]\n",
    "        surr_time = np.array(result[1]['runtime'])[:epochs]\n",
    "        if ug:\n",
    "            runtime = 200*np.array(result[1]['train_time'])[:epochs] + surr_time\n",
    "        else:\n",
    "            runtime = np.array(result[1]['train_time'])[:epochs] + surr_time\n",
    "\n",
    "        val_err = [100 - x for x in val_acc]\n",
    "        val_incumbent = [min(val_err[:epoch]) for epoch in range(1, len(val_err)+1)]\n",
    "        runtime = [sum(runtime[:epoch]) for epoch in range(1, len(runtime)+1)]\n",
    "\n",
    "        if metric == 'valid_acc':\n",
    "            incumbent = val_incumbent\n",
    "        elif metric == 'test_acc':\n",
    "            test_err = [100 - x for x in result[1]['test_acc']]\n",
    "            inc_idx, best, best_idx = [], np.inf, 0\n",
    "            for i, err in enumerate(val_err):\n",
    "                if err < best:\n",
    "                    best, best_idx = err, i\n",
    "                inc_idx.append(best_idx)\n",
    "            incumbent = [test_err[idx] for idx in inc_idx]\n",
    "\n",
    "        if len(incumbent) == epochs:\n",
    "            output.append(incumbent)\n",
    "            time.append(runtime)\n",
    "        else:\n",
    "            nan_count += 1\n",
    "\n",
    "    output = np.array(output)\n",
    "    time = np.array(time)\n",
    "\n",
    "    output, time = get_trajectories(output, time)\n",
    "    \n",
    "    print(predictor, 'output shape', output.shape, 'nans', nan_count, 'missing files', missing_results_count)\n",
    "    mean = np.mean(output, axis=0)\n",
    "    std = np.std(output, axis=0)\n",
    "    std_error = np.sqrt(np.var(output, axis=0, ddof=1)/np.asarray(output).shape[0])\n",
    "    return mean, std, std_error, time\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up colors and plot markings\n",
    "defaults = [(0.12156862745098039, 0.4666666666666667, 0.7058823529411765),\n",
    "            (1.0, 0.4980392156862745, 0.054901960784313725),\n",
    "            (0.17254901960784313, 0.6274509803921569, 0.17254901960784313),\n",
    "            (0.8392156862745098, 0.15294117647058825, 0.1568627450980392),\n",
    "            (0.5803921568627451, 0.403921568627451, 0.7411764705882353),\n",
    "            (0.5490196078431373, 0.33725490196078434, 0.29411764705882354),\n",
    "            (0.8901960784313725, 0.4666666666666667, 0.7607843137254902),\n",
    "            (0.4980392156862745, 0.4980392156862745, 0.4980392156862745),\n",
    "            (0.7372549019607844, 0.7411764705882353, 0.13333333333333333),\n",
    "            (0.09019607843137255, 0.7450980392156863, 0.8117647058823529)]\n",
    "\n",
    "# goes up to 24\n",
    "c_max = 9\n",
    "colors = [*defaults[:c_max], *defaults[:c_max], *defaults[:c_max]]\n",
    "fmts = [*['-']*c_max, *['--']*c_max, *[':']*c_max]\n",
    "markers = [*['^']*c_max, *['v']*c_max, *['o']*c_max]\n",
    "\n",
    "# https://matplotlib.org/2.1.2/api/_as_gen/matplotlib.pyplot.plot.html\n",
    "\n",
    "\n",
    "pred_label_dict={\n",
    "    'valloss':'Val. Loss', 'valacc':'Val. Acc.', 'sotl':'SoTL', 'bananas':'BANANAS',\n",
    "    'mlp':'Feedforward', 'gbdt':'GBDT', 'gcn':'GCN', 'bonas_gcn':'BONAS', 'xgb':'XGB',\n",
    "    'ngb':'NGB', 'rf':'RF', 'jacov':'Jacob. Cov.', 'dngo':'DNGO', 'bohamiann':'BOHAMIANN', \n",
    "    'bayes_lin_reg':'Bayes. Lin. Reg.', 'ff_keras':'FF-Keras', 'gp':'GP', 'sparse_gp':'Sparse GP', \n",
    "    'var_sparse_gp':'Var. Sparse GP', 'seminas':'SemiNAS', 'lcsvr':'LcSVR', 'snip':'SNIP', 'sotle':'SoTLE',\n",
    "    'bonas':'BONAS', 'omni_lofi':'Omni Lofi', 'nao': 'NAO', 'lgb': 'LGB', 'none': 'True'\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mlp output shape (2, 600) nans 1 missing files 0\n",
      "lgb output shape (1, 300) nans 0 missing files 0\n",
      "xgb output shape (2, 600) nans 0 missing files 0\n",
      "rf output shape (2, 600) nans 0 missing files 0\n",
      "bayes_lin_reg output shape (3, 900) nans 0 missing files 0\n",
      "gp output shape (3, 900) nans 0 missing files 0\n",
      "none output shape (3, 900) nans 0 missing files 0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/zelaa/.local/lib/python3.7/site-packages/numpy-1.16.4-py3.7-linux-x86_64.egg/numpy/core/fromnumeric.py:3367: RuntimeWarning: Degrees of freedom <= 0 for slice\n",
      "  **kwargs)\n",
      "/home/zelaa/.local/lib/python3.7/site-packages/numpy-1.16.4-py3.7-linux-x86_64.egg/numpy/core/_methods.py:130: RuntimeWarning: invalid value encountered in true_divide\n",
      "  ret, rcount, out=ret, casting='unsafe', subok=False)\n"
     ]
    }
   ],
   "source": [
    "# set up parameters for the experiments\n",
    "epochs = 300\n",
    "results_dict = {}\n",
    "\n",
    "folder = os.path.expanduser('../../re_run_0/cifar10/nas_predictors/nasbench201')\n",
    "predictors=('mlp', 'lgb', 'xgb', 'rf', 'bayes_lin_reg', 'gp', 'none')\n",
    "\n",
    "for i, predictor in enumerate(predictors):\n",
    "    mean, std, std_error, runtime = get_results(predictor, folder, epochs=epochs, metric='test_acc', ug=True)\n",
    "    results_dict[predictor] = {'label':pred_label_dict[predictor], \n",
    "                               'key':predictor, 'mean':mean, 'std':std, \n",
    "                               'std_error': std_error, 'runtime': runtime}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from mpl_toolkits.axes_grid1.inset_locator import zoomed_inset_axes\n",
    "from mpl_toolkits.axes_grid1.inset_locator import mark_inset\n",
    "from matplotlib.ticker import FormatStrFormatter\n",
    "\n",
    "# plot performance vs runtime\n",
    "\n",
    "# didn't run them long enough to do logspace here. (These experiments took surprisingly long to run)\n",
    "plt.rcParams['axes.grid'] = True\n",
    "plt.rcParams['grid.linestyle'] = 'dotted'\n",
    "\n",
    "plot_zoomed = False\n",
    "plot_sem = True\n",
    "\n",
    "fig, ax = plt.subplots(figsize=[8, 4])\n",
    "if plot_zoomed:\n",
    "    sub_axes = plt.axes([.6, .6, .25, .25]) \n",
    "\n",
    "for i, key in enumerate(predictors):\n",
    "    mean = results_dict[key]['mean']\n",
    "    sem = results_dict[key]['std_error']\n",
    "    label = results_dict[key]['label']\n",
    "    x = results_dict[key]['runtime']\n",
    "    \n",
    "    ax.plot(x, mean, label=label, color=colors[i], linestyle=fmts[i])\n",
    "    if plot_sem:\n",
    "        ax.fill_between(x, mean-1*sem, mean+1*sem,\n",
    "                        color=colors[i], alpha=0.2)\n",
    "    \n",
    "    if plot_zoomed:\n",
    "        X_detail = x[-10000:]\n",
    "        Y_detail = mean[-10000:]\n",
    "        sem_detail = sem[-10000:]\n",
    "        sub_axes.plot(X_detail, Y_detail, color=colors[i], linestyle=fmts[i])\n",
    "        sub_axes.fill_between(X_detail, Y_detail-1*sem_detail, Y_detail+1*sem_detail,\n",
    "                         color=colors[i], alpha=0.2)\n",
    "        sub_axes.tick_params(labelbottom=False) \n",
    "        sub_axes.set_xlim([700000, 900000])\n",
    "    \n",
    "ax.yaxis.set_major_formatter(FormatStrFormatter('%.3f'))\n",
    "ax.set_xscale('log')\n",
    "#ax.set_ylim([7, 11])\n",
    "#ax.set_xlim([1e4, 1.6e6])\n",
    "\n",
    "ax.legend(loc=(1.04,0))\n",
    "ax.set_xlabel('Runtime [s] (simulated)')\n",
    "ax.set_ylabel('Test error (%)')\n",
    "ax.grid(True, which=\"both\",ls=\"-\", alpha=.5)\n",
    "ax.set_title('Test error vs. train time for the Predictor Framework on NAS-Bench-201')\n",
    "plt.savefig('plot_nb201.pdf', bbox_inches = 'tight', pad_inches = 0.1)\n"
   ]
  },
  {
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}