Task_3_ESS.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "colab": {
      "name": "Task_3.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qKo3S_fOsMim"
      },
      "source": [
        "# ML in Cybersecurity: Task 3\n",
        "\n",
        "## Team\n",
        "  * **Team name**: ESS\n",
        "  * **Members**: Eiram Mahera Sheikh (eish00001@stud.uni-saarland.de), Shravan Swaminathan(shsw00001@stud.uni-saarland.de), Shayari Bhattacharjee(shbh00002@stud.uni-saarland.de)\n",
        "\n",
        "\n",
        "## Logistics\n",
        "  * **Due date**: 9th December 2021, 23:59:59\n",
        "  * Email the completed notebook to: `mlcysec_ws2022_staff@lists.cispa.saarland`\n",
        "  * Complete this in **teams of 3**\n",
        "  * Feel free to use the forum to discuss.\n",
        "  \n",
        "## Timeline\n",
        "  * 26-Nov-2021: hand-out\n",
        "  * **09-Dec-2021**: Email completed notebook\n",
        "  \n",
        "  \n",
        "## About this Project\n",
        "In this project, you will explore an application of ML to a popular task in cybersecurity: malware classification.\n",
        "You will be presented with precomputed behaviour analysis reports of thousands of program binaries, many of which are malwares.\n",
        "Your goal is to train a malware detector using this behavioural reports.\n",
        "\n",
        "\n",
        "## A Note on Grading\n",
        "The grading for this project will depend on:\n",
        " 1. Vectorizing Inputs\n",
        "   * Obtaining a reasonable vectorized representations of the input data (a file containing a sequence of system calls)\n",
        "   * Understanding the influence these representations have on your model\n",
        " 1. Classification Model  \n",
        "   * Following a clear ML pipeline\n",
        "   * Obtaining reasonable performances (>60\\%) on held-out test set\n",
        "   * Choice of evaluation metric\n",
        "   * Visualizing loss/accuracy curves\n",
        " 1. Analysis\n",
        "   * Which methods (input representations/ML models) work better than the rest and why?\n",
        "   * Which hyper-parameters and design-choices were important in each of your methods?\n",
        "   * Quantifying influence of these hyper-parameters on loss and/or validation accuracies\n",
        "   * Trade-offs between methods, hyper-parameters, design-choices\n",
        "   * Anything else you find interesting (this part is open-ended)\n",
        "\n",
        "\n",
        "## Grading Details\n",
        " * 40 points: Vectorizing input data (each input = behaviour analysis file in our case)\n",
        " * 40 points: Training a classification model\n",
        " * 15 points: Analysis/Discussion\n",
        " * 5 points: Clean code\n",
        " \n",
        "## Filling-in the Notebook\n",
        "You'll be submitting this very notebook that is filled-in with your code and analysis. Make sure you submit one that has been previously executed in-order. (So that results/graphs are already visible upon opening it). \n",
        "\n",
        "The notebook you submit **should compile** (or should be self-contained and sufficiently commented). Check tutorial 1 on how to set up the Python3 environment.\n",
        "\n",
        "\n",
        "**The notebook is your project report. So, to make the report readable, omit code for techniques/models/things that did not work. You can use the final summary to provide a report about these.**\n",
        "\n",
        "It is extremely important that you **do not** re-order the existing sections. Apart from that, the code blocks that you need to fill-in are given by:\n",
        "```\n",
        "#\n",
        "#\n",
        "# ------- Your Code -------\n",
        "#\n",
        "#\n",
        "```\n",
        "Feel free to break this into multiple-cells. It's even better if you interleave explanations and code-blocks so that the entire notebook forms a readable \"story\".\n",
        "\n",
        "\n",
        "## Code of Honor\n",
        "We encourage discussing ideas and concepts with other students to help you learn and better understand the course content. However, the work you submit and present **must be original** and demonstrate your effort in solving the presented problems. **We will not tolerate** blatantly using existing solutions (such as from the internet), improper collaboration (e.g., sharing code or experimental data between groups) and plagiarism. If the honor code is not met, no points will be awarded.\n",
        "\n",
        " \n",
        " ## Versions\n",
        "  * v1.1: Updated deadline\n",
        "  * v1.0: Initial notebook\n",
        "  \n",
        "  ---"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# !pip install numpy\n",
        "# !pip install matplotlib\n",
        "# !pip install json\n",
        "# !pip install tqdm\n",
        "# !pip install torch\n",
        "# !pip install pandas\n",
        "# !pip install keras\n",
        "# !pip install nltk\n",
        "# !pip install tensorflow\n"
      ],
      "metadata": {
        "id": "2924tdH9QYWs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "uyCiLbXbsMiq"
      },
      "source": [
        "import time \n",
        "import gc\n",
        " \n",
        "import numpy as np \n",
        "import matplotlib.pyplot as plt \n",
        "\n",
        "import json \n",
        "import time \n",
        "import pickle \n",
        "import sys \n",
        "import csv \n",
        "import os \n",
        "import os.path as osp \n",
        "import shutil \n",
        "import pathlib\n",
        "from pathlib import Path\n",
        "\n",
        "from IPython.display import display, HTML\n",
        " \n",
        "%matplotlib inline \n",
        "plt.rcParams['figure.figsize'] = (10.0, 8.0) # set default size of plots \n",
        "plt.rcParams['image.interpolation'] = 'nearest' \n",
        "plt.rcParams['image.cmap'] = 'gray' \n",
        " \n",
        "# for auto-reloading external modules \n",
        "# see http://stackoverflow.com/questions/1907993/autoreload-of-modules-in-ipython \n",
        "%load_ext autoreload\n",
        "%autoreload 2"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "7dHogdxRsMis"
      },
      "source": [
        "# Some suggestions of our libraries that might be helpful for this project\n",
        "from collections import Counter          # an even easier way to count\n",
        "from multiprocessing import Pool         # for multiprocessing\n",
        "from tqdm import tqdm                    # fancy progress bars\n",
        "\n",
        "# Load other libraries here.\n",
        "# Keep it minimal! We should be easily able to reproduce your code.\n",
        "\n",
        "# We preload pytorch as an example\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import pandas\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "from torch.utils.data import DataLoader, Dataset, TensorDataset\n",
        "\n",
        "from keras.preprocessing.text import text_to_word_sequence\n",
        "from nltk import ngrams"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oJWkh3GUsMit"
      },
      "source": [
        "# Setup\n",
        "\n",
        "  * Download the datasets: [train](https://nextcloud.mpi-klsb.mpg.de/index.php/s/pJrRGzm2So2PMZm) (128M) and [test](https://nextcloud.mpi-klsb.mpg.de/index.php/s/zN3yeWzQB3i5WqE) (92M)\n",
        "  * Unpack them under `./data/train` and `./data/test`\n",
        "  * Hint: you can execute shell scripts from notebooks using the `!` prefix, e.g., `! wget <url>`"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# !pip install google.colab"
      ],
      "metadata": {
        "id": "qBNy6nZPT73e"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KlmVxND76Hgo",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e4c774b9-e8e5-43b6-e797-c96ce4ad921e"
      },
      "source": [
        "# We have stored the data on google drive for persistence.\n",
        "# Mount the drive\n",
        "\n",
        "from google.colab import drive\n",
        "drive.mount('/content/drive', force_remount=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/drive\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ezhb2E_RTq1O"
      },
      "source": [
        "train_path = \"drive/MyDrive/data/data/train\"\n",
        "test_path = \"drive/MyDrive/data/data/test\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wlhs4w44sMit",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b14676d6-0e6b-42fa-ee98-687839f715b8"
      },
      "source": [
        "# Check that you are prepared with the data\n",
        "! printf '# train examples (Should be 13682) : '; ls {train_path} | wc -l\n",
        "! printf '# test  examples (Should be 10000) : '; ls {test_path} | wc -l"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "# train examples (Should be 13682) : 13692\n",
            "# test  examples (Should be 10000) : 10020\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Zh1-JHNIsMiv"
      },
      "source": [
        "Now that you're set, let's briefly look at the data you have been handed.\n",
        "Each file encodes the behavior report of a program (potentially a malware), using an encoding scheme called \"The Malware Instruction Set\" (MIST for short).\n",
        "At this point, we highly recommend you briefly read-up Sec. 2 of the [MIST](http://www.mlsec.org/malheur/docs/mist-tr.pdf) documentation.\n",
        "\n",
        "You will find each file named as `filename.<malwarename>`:\n",
        "```\n",
        "» ls data/train | head\n",
        "00005ecc06ae3e489042e979717bb1455f17ac9d.NothingFound\n",
        "0008e3d188483aeae0de62d8d3a1479bd63ed8c9.Basun\n",
        "000d2eea77ee037b7ef99586eb2f1433991baca9.Patched\n",
        "000d996fa8f3c83c1c5568687bb3883a543ec874.Basun\n",
        "0010f78d3ffee61101068a0722e09a98959a5f2c.Basun\n",
        "0013cd0a8febd88bfc4333e20486bd1a9816fcbf.Basun\n",
        "0014aca72eb88a7f20fce5a4e000c1f7fff4958a.Texel\n",
        "001ffc75f24a0ae63a7033a01b8152ba371f6154.Texel\n",
        "0022d6ba67d556b931e3ab26abcd7490393703c4.Basun\n",
        "0028c307a125cf0fdc97d7a1ffce118c6e560a70.Swizzor\n",
        "...\n",
        "```\n",
        "and within each file, you will see a sequence of individual systems calls monitored duing the run-time of the binary - a malware named 'Basun' in the case:\n",
        "```\n",
        "» head data/train/000d996fa8f3c83c1c5568687bb3883a543ec874.Basun\n",
        "# process 000006c8 0000066a 022c82f4 00000000 thread 0001 #\n",
        "02 01 | 000006c8 0000066a 00015000\n",
        "02 02 | 00006b2c 047c8042 000b9000\n",
        "02 02 | 00006b2c 047c8042 00108000\n",
        "02 02 | 00006b2c 047c8042 00153000\n",
        "02 02 | 00006b2c 047c8042 00091000\n",
        "02 02 | 00006b2c 047c8042 00049000\n",
        "02 02 | 00006b2c 047c8042 000aa000\n",
        "02 02 | 00006b2c 047c8042 00092000\n",
        "02 02 | 00006b2c 047c8042 00011000\n",
        "...\n",
        "```\n",
        "(**Note**: Please ignore the first line that begins with `# process ...`.)\n",
        "\n",
        "Your task in this project is to train a malware detector, which given the sequence of system calls (in the MIST-formatted file like above), predicts one of 10 classes: `{ Agent, Allaple, AutoIt, Basun, NothingFound, Patched, Swizzor, Texel, VB, Virut }`, where `NothingFound` roughly represents no malware is present.\n",
        "In terms of machine learning terminology, your malware detector $F: X \\rightarrow Y$ should learn a mapping from the MIST-encoded behaviour report (the input $x \\in X$) to the malware class $y \\in Y$.\n",
        "\n",
        "Consequently, you will primarily tackle two challenges in this project:\n",
        "  1. \"Vectorizing\" the input data i.e., representing each input (file) as a tensor\n",
        "  1. Training an ML model\n",
        "  \n",
        "\n",
        "### Some tips:\n",
        "  * Begin with an extremely simple representation/ML model and get above chance-level classification performance\n",
        "  * Choose your evaluation metric wisely\n",
        "  * Save intermediate computations (e.g., a token to index mapping). This will avoid you parsing the entire dataset for every experiment\n",
        "  * Try using `multiprocessing.Pool` to parallelize your `for` loops"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p8xloR8dsMiv"
      },
      "source": [
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vXnyDAdbsMiw"
      },
      "source": [
        "# 1. Vectorize Data"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ia-xFIQtsMiw"
      },
      "source": [
        "## 1.a. Load Raw Data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "ZScY-C2IsMiw"
      },
      "source": [
        "def load_content(filepath):\n",
        "    '''Given a filepath, returns (content, classname), where content = [list of lines in file]'''\n",
        "    label = filepath.split('.')[-1]\n",
        "    \n",
        "    # open the file within a context manager and read all the lines\n",
        "    with open(filepath, \"r\") as fh:\n",
        "      lines = fh.readlines()\n",
        "    \n",
        "    return lines, label\n",
        "\n",
        "\n",
        "def load_data(data_path, nworkers=10):\n",
        "    '''Returns each data sample as a tuple (x, y), x = sequence of strings (i.e., syscalls), y = malware program class'''\n",
        "    raw_data_samples = []\n",
        "    \n",
        "    # get a list of all files in the training data\n",
        "    files = os.listdir(data_path)\n",
        "\n",
        "    # load the content of each file\n",
        "    for file in tqdm(files):\n",
        "        lines, label = load_content(os.path.join(data_path, file))\n",
        "        raw_data_samples.append((lines, label))\n",
        "\n",
        "    return raw_data_samples"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "7BSUdq_-sMiw",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8c5ce0df-8477-4369-bfb3-e3b9b079c6d7"
      },
      "source": [
        "print('=> Loading training data ... ')\n",
        "train_raw_samples = load_data(Path(train_path))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "=> Loading training data ... \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "  3%|▎         | 399/13692 [01:43<57:24,  3.86it/s]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "GG6tcSNJsMiw",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5185baa7-c73a-4a7d-f410-01463c4a22d1"
      },
      "source": [
        "project_mode = 'trainval'    # trainval, traintest, debug\n",
        "np.random.seed(123)          # To perform the same split across multiple runs\n",
        "\n",
        "if project_mode == 'trainval':\n",
        "    # Split data into train and validation set\n",
        "    train_size = int(len(train_raw_samples) * 0.9)\n",
        "    np.random.shuffle(train_raw_samples)\n",
        "    train_raw_samples, test_raw_samples = train_raw_samples[:train_size], train_raw_samples[train_size:]\n",
        "\n",
        "elif project_mode == 'traintest':\n",
        "    # Load train and test data\n",
        "    test_raw_samples = load_data(Path(test_path))\n",
        "\n",
        "elif project_mode == 'debug':\n",
        "    np.random.shuffle(train_raw_samples)\n",
        "    train_raw_samples, test_raw_samples = train_raw_samples[:1000], train_raw_samples[1001:1250]\n",
        "\n",
        "else:\n",
        "    raise ValueError('Unrecognized mode')\n",
        "    \n",
        "print('=> # Train samples = ', len(train_raw_samples))\n",
        "print('=> # Test  samples = ', len(test_raw_samples))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "=> # Train samples =  360\n",
            "=> # Test  samples =  40\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GjiNiOVHsMiy"
      },
      "source": [
        "## 1.b. Vectorize: Setup\n",
        "\n",
        "Make one pass over the inputs to identify relevant features/tokens.\n",
        "\n",
        "Suggestion:\n",
        "  - identify tokens (e.g., unigrams, bigrams)\n",
        "  - create a token -> index (int) mapping. Note that you might have a >10K unique tokens. So, you will have to choose a suitable \"vocabulary\" size."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "DAtmzL6PsMiy"
      },
      "source": [
        "class_to_idx = {\n",
        "    'Agent':0, \n",
        "    'Allaple':1, \n",
        "    'AutoIt':2, \n",
        "    'Basun':3, \n",
        "    'NothingFound':4, \n",
        "    'Patched':5, \n",
        "    'Swizzor':6, \n",
        "    'Texel':7, \n",
        "    'VB':8, \n",
        "    'Virut':9\n",
        "    }\n",
        "\n",
        "   \n",
        "def get_key_idx_map(counter, vocab_size, ukn_token='_ukn_'):\n",
        "    key_to_idx, idx_to_key = dict(), dict()\n",
        "    word_count = list(counter.most_common(vocab_size))\n",
        "    counter = 0\n",
        "    \n",
        "    for word, count in word_count:\n",
        "        key_to_idx[word] = counter\n",
        "        idx_to_key[counter] = word\n",
        "        counter += 1\n",
        "\n",
        "    return key_to_idx, idx_to_key\n",
        "\n",
        "\n",
        "def helper1_tokenize(raw_data):\n",
        "    tokenized_data = []\n",
        "    for content, label in raw_data:\n",
        "        tokens = []\n",
        "        for line in content[1: -1]:\n",
        "            tokens.extend(text_to_word_sequence(line))\n",
        "        tokenized_data.append(tokens)\n",
        "\n",
        "    return tokenized_data\n",
        "\n",
        "\n",
        "def helper2_ngrams(tokens, n):\n",
        "    n_grams = []\n",
        "    for token_set in tokens:\n",
        "        n_grams.append(list(ngrams(token_set, n)))\n",
        "\n",
        "    return n_grams\n",
        "\n",
        "\n",
        "def helper3_get_class_to_idx_map(raw_data):\n",
        "    labels_y = list()\n",
        "    for counter, labels in raw_data:\n",
        "        labels_y.append(class_to_idx[labels])\n",
        "\n",
        "    return np.array(labels_y)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "LSMEAcgvsMiy"
      },
      "source": [
        "gc.collect()\n",
        "\n",
        "tokens = helper1_tokenize(train_raw_samples)\n",
        "\n",
        "#unigrams = helper2_ngrams(tokens, 1) #for choosing models of unigram-bigrams or trigrams\n",
        "bigrams = helper2_ngrams(tokens, 2)\n",
        "#trigrams = helper2_ngrams(tokens, 3)\n",
        "\n",
        "tokens = []\n",
        "\n",
        "MAX_VOCAB_SIZE = 300\n",
        "\n",
        "ngram_tokens = []\n",
        "for ngram_set in bigrams:\n",
        "    ngram_tokens.extend(ngram_set)\n",
        "\n",
        "token_to_idx, idx_to_token = get_key_idx_map(Counter(ngram_tokens), MAX_VOCAB_SIZE)\n",
        "ngram_tokens = []\n",
        "train_y = helper3_get_class_to_idx_map(train_raw_samples)\n",
        "test_y = helper3_get_class_to_idx_map(test_raw_samples)\n",
        "\n",
        "# Save vocab to file\n",
        "out_path = 'application_vocab_{}.pkl'.format(MAX_VOCAB_SIZE)\n",
        "with open(out_path, 'wb') as wf:\n",
        "    dct = {\n",
        "        'token_to_idx': token_to_idx,\n",
        "        'idx_to_token': idx_to_token\n",
        "        }\n",
        "    pickle.dump(dct, wf)\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QUefzVjFsMiy"
      },
      "source": [
        "## 1.c. Vectorize Data\n",
        "\n",
        "Use the (token $\\rightarrow$ index) mapping you created before to vectorize your data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "ffJJ0XOvsMiz"
      },
      "source": [
        "gc.collect()\n",
        "\n",
        "def BinaryVectorizer(tokens):\n",
        "    vectorized = []\n",
        "\n",
        "    for token in tokens:\n",
        "        vector = np.zeros(MAX_VOCAB_SIZE)\n",
        "        for word in token:\n",
        "            if word in token_to_idx:\n",
        "                vector[token_to_idx[word]] = 1\n",
        "        vectorized.append(np.array(vector, dtype='float32'))\n",
        "\n",
        "    return np.array(vectorized, dtype='float32')\n",
        "\n",
        "\n",
        "def CountVectorizer(ngrams):\n",
        "    vectorized = []\n",
        "\n",
        "    for ng in ngrams:\n",
        "        vector = np.zeros(MAX_VOCAB_SIZE)\n",
        "        token_counter = dict(Counter(ng))\n",
        "        unique_tokens = list(token_counter.keys())\n",
        "        for token in unique_tokens:\n",
        "            if token in token_to_idx:\n",
        "                vector[token_to_idx[token]] = token_counter[token]\n",
        "        vectorized.append(np.array(vector, dtype='float32'))\n",
        "\n",
        "    return np.array(vectorized, dtype='float32')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "CD5IqJmXsMiz"
      },
      "source": [
        "def vectorize_raw_samples(raw_samples, nworkers=10):\n",
        "    vectorized_samples = []\n",
        "    \n",
        "    tokens = helper1_tokenize(raw_samples)\n",
        "    ngram_data = helper2_ngrams(tokens, 2)\n",
        "    vectorized_samples = BinaryVectorizer(ngram_data)\n",
        "    #vectorized_samples = CountVectorizer(ngram_data) #for models utilising count vectorizer\n",
        "\n",
        "    return vectorized_samples"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "scrolled": true,
        "id": "BtS10ASbsMi0",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5d9cf6b6-0ab1-4307-e63e-e5e9cc79aea6"
      },
      "source": [
        "print('=> Processing: Train')\n",
        "train_data = vectorize_raw_samples(train_raw_samples)\n",
        "print()\n",
        "print('=> Processing: Test')\n",
        "test_data = vectorize_raw_samples(test_raw_samples)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "=> Processing: Train\n",
            "\n",
            "=> Processing: Test\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LDtbwCthNM4V"
      },
      "source": [
        "train_x = torch.from_numpy(train_data).float()\n",
        "test_x = torch.from_numpy(test_data).float()\n",
        "\n",
        "train_y = torch.tensor(train_y, dtype=torch.long)\n",
        "test_y = torch.tensor(test_y, dtype=torch.long)\n",
        "\n",
        "trainset = TensorDataset(train_x, train_y)\n",
        "testset = TensorDataset(test_x, test_y)\n",
        "\n",
        "torch.save(trainset, 'trainset.pt')\n",
        "torch.save(testset, 'testset.pt')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_l0tLsEVsMi0"
      },
      "source": [
        "# 2. Train Model\n",
        "\n",
        "You will now train an ML model on the vectorized datasets you created previously.\n",
        "\n",
        "_Note_: Although we often refer to each input as a 'vector' for simplicity, each of your inputs can also be higher dimensional tensors."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UInY1YzksMi0"
      },
      "source": [
        "## 2.a. Helpers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "VnSv0J75sMi0"
      },
      "source": [
        "def evaluate_preds(y_predicted,y_true):\n",
        "    precision = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    recall = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    positive = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    true_positive = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    f1 = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0} \n",
        "    classes = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    \n",
        "    true_samples = generate_true_labels(y_true)\n",
        "    for out_idx in range(len(y_predicted)):\n",
        "        for inner_idx in range(len(y_predicted[out_idx])):\n",
        "            if y_predicted[out_idx][inner_idx] == y_true[out_idx][inner_idx]:\n",
        "                true_positive[y_predicted[out_idx][inner_idx]] += 1\n",
        "            positive[y_predicted[out_idx][inner_idx]] += 1\n",
        "            classes[y_predicted[out_idx][inner_idx]] += 1\n",
        "    \n",
        "    for key, vals in classes.items():\n",
        "        if positive[key] <= 0.0:\n",
        "            precision[key] = 0.0\n",
        "        else:\n",
        "            precision[key] = (true_positive[key]/positive[key]) * 100\n",
        "        if true_samples[key] <= 0.0:\n",
        "            recall[key] = 0.0\n",
        "        else:\n",
        "            recall[key] = (true_positive[key]/true_samples[key]) * 100\n",
        "    \n",
        "    for key, vals in classes.items():\n",
        "        if recall[key] <= 0.0:\n",
        "            f1[key] = 0.0\n",
        "        else:\n",
        "            f1[key] = 2*((precision[key]*recall[key])/(precision[key]+recall[key]))\n",
        "    return precision, recall, f1\n",
        "    pass\n",
        "\n",
        "\n",
        "def generate_true_labels(y_true):\n",
        "    true_labels = {0:0,1:0,2:0,3:0,4:0,5:0,6:0,7:0,8:0,9:0}\n",
        "    for labels in y_true:\n",
        "        for val in labels:\n",
        "            true_labels[val] += 1\n",
        "    return true_labels\n",
        "\n",
        "def visualize_evaluation(classes, precision, recall, f1):\n",
        "    data = {'Labels/Malwares': list(classes.keys()), 'Precision': list(precision.values()), 'Recall': list(recall.values()), 'F1 Score': list(f1.values())}\n",
        "    dataframe = pandas.DataFrame(data=data)\n",
        "    print(dataframe)\n",
        "\n",
        "\n",
        "def save_model(model, path):\n",
        "    checkpoint = {'epoch': epoch,'state_dict': model.state_dict(),'optimizer' : optimizer.state_dict(),'loss': loss_function}\n",
        "    torch.save(checkpoint, path)\n",
        "    pass\n",
        "\n",
        "\n",
        "def save_data(eval_data, path):\n",
        "    with open(path, 'wb') as wf:\n",
        "        pickle.dump(eval_data, wf)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PhbVzVLDsMi1"
      },
      "source": [
        "## 2.b. Define Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jrUKCNRgsMi1"
      },
      "source": [
        "Describe your model here."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "ILB2fBRWsMi1"
      },
      "source": [
        "class NetTemplate1(nn.Module):\n",
        "    def __init__(self, input_dim, h1, h2, h3, output_dim):\n",
        "        super(NetTemplate1, self).__init__()\n",
        "        self.fc1 = nn.Linear(input_dim, h1)\n",
        "        self.fc2 = nn.Linear(h1, h2)\n",
        "        self.fc3 = nn.Linear(h2, h3)\n",
        "        self.fc4 = nn.Linear(h3, output_dim)\n",
        "\n",
        "    def forward(self, x):\n",
        "        # Forward pass\n",
        "        x = F.relu(self.fc1(x))\n",
        "        x = F.relu(self.fc2(x))\n",
        "        x = F.relu(self.fc3(x))\n",
        "        x = self.fc4(x)\n",
        "        return F.log_softmax(x)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QO1yPx7YsMi1"
      },
      "source": [
        "## 2.c. Set Hyperparameters"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "0k0jLX5ksMi1"
      },
      "source": [
        "# Define your hyperparameters here\n",
        "\n",
        "in_dims = trainset[0][0].shape[0]\n",
        "out_dims = len(class_to_idx)\n",
        "h1, h2, h3 = 128, 128, 64\n",
        "out_dims = len(class_to_idx)\n",
        "\n",
        "# Optimization\n",
        "n_epochs = 50\n",
        "batch_size=16\n",
        "batch_size_train = 64\n",
        "batch_size_test = 1000\n",
        "learning_rate = 0.01\n",
        "momentum = 0.5\n",
        "log_interval = 10"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NtpgVAO1sMi2"
      },
      "source": [
        "## 2.d. Train your Model"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "compute_mode = 'gpu'\n",
        "\n",
        "if compute_mode == 'cpu':\n",
        "    device = torch.device('cpu')\n",
        "elif compute_mode == 'gpu':\n",
        "    os.environ[\"CUDA_VISIBLE_DEVICES\"] = '0'    # Set device ID here\n",
        "    device = torch.device('cuda')\n",
        "else:\n",
        "    raise ValueError('Unrecognized compute mode')"
      ],
      "metadata": {
        "id": "u6e0EXJYGFYF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "gc.collect()\n",
        "\n",
        "train_x = torch.from_numpy(train_data).float().to(device)\n",
        "test_x = torch.from_numpy(test_data).float().to(device)\n",
        "\n",
        "train_y = torch.tensor(train_y, dtype=torch.long).to(device)\n",
        "test_y = torch.tensor(test_y, dtype=torch.long).to(device)\n",
        "\n",
        "trainset = TensorDataset(train_x, train_y)\n",
        "testset = TensorDataset(test_x, test_y)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Vi5JK1sFHo6T",
        "outputId": "07cf3758-194e-456c-b684-9f9ed444543f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:6: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
            "  \n",
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:7: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
            "  import sys\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "collapsed": true,
        "id": "Co0W2ifGsMi2"
      },
      "source": [
        "network = NetTemplate1(in_dims, h1, h2, h3, out_dims).to(device)\n",
        "\n",
        "optimizer = optim.SGD(network.parameters(), lr=learning_rate,momentum=momentum)\n",
        "#optimzer = optim.AdamW(params, lr=learning_rate, betas=(0.9, 0.999), eps=1e-08, weight_decay=0.01, amsgrad=False) #for models using AdamW as optimiser\n",
        "loss_function = nn.NLLLoss()\n",
        "\n",
        "train_loader = DataLoader(dataset = trainset, batch_size=batch_size, shuffle=True)\n",
        "test_loader = DataLoader(dataset = testset, batch_size=batch_size, shuffle=False)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "predictions = []\n",
        "targets = []\n",
        "train_losses = []\n",
        "train_counter = []\n",
        "test_losses = []\n",
        "test_counter = [i*len(test_loader.dataset) for i in range(0,n_epochs)]\n",
        "test_accuracies = []\n",
        "\n",
        "\n",
        "def train(epoch):\n",
        "    for batch_idx, (data, target) in enumerate(train_loader):\n",
        "        data, target = data.to(device),target.to(device)\n",
        "        data = data.view(-1, in_dims)\n",
        "        optimizer.zero_grad()\n",
        "        net_out = network(data)\n",
        "        loss = loss_function(net_out, target)\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "        train_losses.append(loss.item())\n",
        "        train_counter.append((batch_idx*64) + ((epoch-1)*len(train_loader.dataset)))\n",
        "        if batch_idx % log_interval == 0:\n",
        "            print('BIGRAM-MLP-Binary-SGD: Train Epoch: {} [{}/{} ({:.0f}%)]\\tLoss: {:.6f}'.format(\n",
        "                epoch, batch_idx * len(data), len(train_loader.dataset),\n",
        "                    100. * batch_idx / len(train_loader), loss.item()))\n",
        "            \n",
        "def test():    \n",
        "    loss = 0\n",
        "    correct = 0\n",
        "    \n",
        "    for data, target in test_loader:\n",
        "        data, target = data.to(device), target.to(device)\n",
        "        data = data.view(-1, in_dims)\n",
        "        net_out = network(data)\n",
        "        loss += loss_function(net_out, target).item()\n",
        "        pred = net_out.data.max(1)[1]\n",
        "        predictions.append(pred.data.to('cpu').numpy())\n",
        "        targets.append(target.data.to('cpu').numpy())\n",
        "        correct += pred.eq(target.data).sum()\n",
        "\n",
        "    loss /= len(test_loader.dataset)\n",
        "    test_accuracies.append(100 * (correct / len(test_loader.dataset)))\n",
        "    test_losses.append(loss)\n",
        "    print('\\n BIGRAM-MLP-Binary-SGD: Test set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\\n'.format(\n",
        "            loss, correct, len(test_loader.dataset),\n",
        "            100. * correct / len(test_loader.dataset)))"
      ],
      "metadata": {
        "id": "sKH7BtzRB6kw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yYipW-8GDREU",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cb4ead74-0e7a-4246-da59-23791060bd4e"
      },
      "source": [
        "for epoch in range(1, n_epochs+1):\n",
        "    train(epoch)\n",
        "    test()\n",
        "    "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:15: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
            "  from ipykernel import kernelapp as app\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 1 [0/360 (0%)]\tLoss: 2.288965\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 1 [160/360 (43%)]\tLoss: 2.275971\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 1 [320/360 (87%)]\tLoss: 2.222792\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1666, Accuracy: 14/40 (35%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 2 [0/360 (0%)]\tLoss: 2.224549\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 2 [160/360 (43%)]\tLoss: 2.211091\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 2 [320/360 (87%)]\tLoss: 2.236853\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1620, Accuracy: 14/40 (35%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 3 [0/360 (0%)]\tLoss: 2.171165\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 3 [160/360 (43%)]\tLoss: 2.235524\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 3 [320/360 (87%)]\tLoss: 2.149633\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1573, Accuracy: 14/40 (35%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 4 [0/360 (0%)]\tLoss: 2.154296\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 4 [160/360 (43%)]\tLoss: 2.085789\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 4 [320/360 (87%)]\tLoss: 2.012789\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1520, Accuracy: 14/40 (35%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 5 [0/360 (0%)]\tLoss: 2.145101\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 5 [160/360 (43%)]\tLoss: 2.025685\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 5 [320/360 (87%)]\tLoss: 2.193282\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1462, Accuracy: 14/40 (35%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 6 [0/360 (0%)]\tLoss: 2.186268\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 6 [160/360 (43%)]\tLoss: 2.118084\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 6 [320/360 (87%)]\tLoss: 1.832987\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1400, Accuracy: 19/40 (48%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 7 [0/360 (0%)]\tLoss: 1.828515\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 7 [160/360 (43%)]\tLoss: 1.856034\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 7 [320/360 (87%)]\tLoss: 2.032290\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1339, Accuracy: 20/40 (50%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 8 [0/360 (0%)]\tLoss: 1.512210\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 8 [160/360 (43%)]\tLoss: 1.600670\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 8 [320/360 (87%)]\tLoss: 1.608413\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1280, Accuracy: 20/40 (50%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 9 [0/360 (0%)]\tLoss: 1.854942\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 9 [160/360 (43%)]\tLoss: 1.676322\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 9 [320/360 (87%)]\tLoss: 1.444990\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1221, Accuracy: 20/40 (50%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 10 [0/360 (0%)]\tLoss: 1.809559\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 10 [160/360 (43%)]\tLoss: 1.983765\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 10 [320/360 (87%)]\tLoss: 1.843540\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1156, Accuracy: 20/40 (50%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 11 [0/360 (0%)]\tLoss: 1.809096\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 11 [160/360 (43%)]\tLoss: 1.865740\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 11 [320/360 (87%)]\tLoss: 1.636263\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1101, Accuracy: 20/40 (50%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 12 [0/360 (0%)]\tLoss: 1.419712\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 12 [160/360 (43%)]\tLoss: 1.093553\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 12 [320/360 (87%)]\tLoss: 1.655333\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.1038, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 13 [0/360 (0%)]\tLoss: 1.323976\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 13 [160/360 (43%)]\tLoss: 1.219160\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 13 [320/360 (87%)]\tLoss: 1.343149\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0997, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 14 [0/360 (0%)]\tLoss: 1.086664\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 14 [160/360 (43%)]\tLoss: 1.622938\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 14 [320/360 (87%)]\tLoss: 1.434181\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0929, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 15 [0/360 (0%)]\tLoss: 1.440659\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 15 [160/360 (43%)]\tLoss: 1.177480\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 15 [320/360 (87%)]\tLoss: 1.742770\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0904, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 16 [0/360 (0%)]\tLoss: 1.177545\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 16 [160/360 (43%)]\tLoss: 1.259587\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 16 [320/360 (87%)]\tLoss: 0.897235\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0862, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 17 [0/360 (0%)]\tLoss: 1.446965\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 17 [160/360 (43%)]\tLoss: 0.794244\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 17 [320/360 (87%)]\tLoss: 1.165253\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0819, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 18 [0/360 (0%)]\tLoss: 0.956615\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 18 [160/360 (43%)]\tLoss: 1.084290\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 18 [320/360 (87%)]\tLoss: 1.095238\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0809, Accuracy: 24/40 (60%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 19 [0/360 (0%)]\tLoss: 0.931092\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 19 [160/360 (43%)]\tLoss: 1.244163\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 19 [320/360 (87%)]\tLoss: 1.024723\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0773, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 20 [0/360 (0%)]\tLoss: 1.305277\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 20 [160/360 (43%)]\tLoss: 1.173761\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 20 [320/360 (87%)]\tLoss: 1.250224\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0727, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 21 [0/360 (0%)]\tLoss: 1.150767\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 21 [160/360 (43%)]\tLoss: 1.295751\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 21 [320/360 (87%)]\tLoss: 1.202646\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0724, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 22 [0/360 (0%)]\tLoss: 1.226592\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 22 [160/360 (43%)]\tLoss: 1.411211\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 22 [320/360 (87%)]\tLoss: 0.898115\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0726, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 23 [0/360 (0%)]\tLoss: 1.310245\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 23 [160/360 (43%)]\tLoss: 0.677483\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 23 [320/360 (87%)]\tLoss: 0.990454\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0675, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 24 [0/360 (0%)]\tLoss: 1.170449\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 24 [160/360 (43%)]\tLoss: 0.754761\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 24 [320/360 (87%)]\tLoss: 0.817507\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0678, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 25 [0/360 (0%)]\tLoss: 0.805065\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 25 [160/360 (43%)]\tLoss: 0.752278\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 25 [320/360 (87%)]\tLoss: 1.082775\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0654, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 26 [0/360 (0%)]\tLoss: 0.714420\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 26 [160/360 (43%)]\tLoss: 0.700892\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 26 [320/360 (87%)]\tLoss: 0.820009\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0643, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 27 [0/360 (0%)]\tLoss: 0.862469\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 27 [160/360 (43%)]\tLoss: 0.677574\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 27 [320/360 (87%)]\tLoss: 1.059058\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0633, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 28 [0/360 (0%)]\tLoss: 0.669210\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 28 [160/360 (43%)]\tLoss: 0.706541\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 28 [320/360 (87%)]\tLoss: 1.053044\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0607, Accuracy: 30/40 (75%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 29 [0/360 (0%)]\tLoss: 0.665593\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 29 [160/360 (43%)]\tLoss: 0.804427\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 29 [320/360 (87%)]\tLoss: 1.350638\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0644, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 30 [0/360 (0%)]\tLoss: 1.066888\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 30 [160/360 (43%)]\tLoss: 0.907245\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 30 [320/360 (87%)]\tLoss: 0.996750\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0637, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 31 [0/360 (0%)]\tLoss: 0.955448\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 31 [160/360 (43%)]\tLoss: 0.751373\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 31 [320/360 (87%)]\tLoss: 0.924962\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0600, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 32 [0/360 (0%)]\tLoss: 0.982418\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 32 [160/360 (43%)]\tLoss: 0.932833\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 32 [320/360 (87%)]\tLoss: 0.793836\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0636, Accuracy: 29/40 (72%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 33 [0/360 (0%)]\tLoss: 1.038621\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 33 [160/360 (43%)]\tLoss: 1.276794\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 33 [320/360 (87%)]\tLoss: 0.850890\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0573, Accuracy: 27/40 (68%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 34 [0/360 (0%)]\tLoss: 0.790399\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 34 [160/360 (43%)]\tLoss: 0.521191\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 34 [320/360 (87%)]\tLoss: 1.254526\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0573, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 35 [0/360 (0%)]\tLoss: 0.652192\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 35 [160/360 (43%)]\tLoss: 0.900999\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 35 [320/360 (87%)]\tLoss: 0.940092\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0570, Accuracy: 31/40 (78%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 36 [0/360 (0%)]\tLoss: 0.824668\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 36 [160/360 (43%)]\tLoss: 0.719681\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 36 [320/360 (87%)]\tLoss: 0.386738\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0622, Accuracy: 30/40 (75%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 37 [0/360 (0%)]\tLoss: 0.762281\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 37 [160/360 (43%)]\tLoss: 0.655580\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 37 [320/360 (87%)]\tLoss: 0.816066\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0578, Accuracy: 31/40 (78%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 38 [0/360 (0%)]\tLoss: 1.013443\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 38 [160/360 (43%)]\tLoss: 0.399309\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 38 [320/360 (87%)]\tLoss: 0.714902\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0523, Accuracy: 33/40 (82%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 39 [0/360 (0%)]\tLoss: 0.714342\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 39 [160/360 (43%)]\tLoss: 0.503329\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 39 [320/360 (87%)]\tLoss: 0.911154\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0561, Accuracy: 32/40 (80%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 40 [0/360 (0%)]\tLoss: 0.786677\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 40 [160/360 (43%)]\tLoss: 0.859033\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 40 [320/360 (87%)]\tLoss: 0.637652\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0562, Accuracy: 32/40 (80%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 41 [0/360 (0%)]\tLoss: 1.130699\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 41 [160/360 (43%)]\tLoss: 0.379374\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 41 [320/360 (87%)]\tLoss: 0.513435\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0600, Accuracy: 30/40 (75%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 42 [0/360 (0%)]\tLoss: 0.666836\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 42 [160/360 (43%)]\tLoss: 0.589085\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 42 [320/360 (87%)]\tLoss: 0.733115\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0563, Accuracy: 31/40 (78%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 43 [0/360 (0%)]\tLoss: 0.799612\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 43 [160/360 (43%)]\tLoss: 0.239570\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 43 [320/360 (87%)]\tLoss: 0.892979\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0596, Accuracy: 31/40 (78%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 44 [0/360 (0%)]\tLoss: 0.599659\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 44 [160/360 (43%)]\tLoss: 0.703742\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 44 [320/360 (87%)]\tLoss: 0.430587\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0606, Accuracy: 29/40 (72%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 45 [0/360 (0%)]\tLoss: 0.513570\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 45 [160/360 (43%)]\tLoss: 0.458826\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 45 [320/360 (87%)]\tLoss: 0.552963\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0572, Accuracy: 33/40 (82%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 46 [0/360 (0%)]\tLoss: 0.664840\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 46 [160/360 (43%)]\tLoss: 0.733628\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 46 [320/360 (87%)]\tLoss: 1.147728\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0596, Accuracy: 29/40 (72%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 47 [0/360 (0%)]\tLoss: 0.554616\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 47 [160/360 (43%)]\tLoss: 0.727493\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 47 [320/360 (87%)]\tLoss: 0.875832\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0590, Accuracy: 30/40 (75%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 48 [0/360 (0%)]\tLoss: 0.665202\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 48 [160/360 (43%)]\tLoss: 0.426986\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 48 [320/360 (87%)]\tLoss: 0.471412\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0589, Accuracy: 30/40 (75%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 49 [0/360 (0%)]\tLoss: 0.793057\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 49 [160/360 (43%)]\tLoss: 0.515556\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 49 [320/360 (87%)]\tLoss: 0.508227\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0636, Accuracy: 28/40 (70%)\n",
            "\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 50 [0/360 (0%)]\tLoss: 0.573429\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 50 [160/360 (43%)]\tLoss: 0.464337\n",
            "BIGRAM-MLP-Binary-SGD: Train Epoch: 50 [320/360 (87%)]\tLoss: 0.566752\n",
            "\n",
            " BIGRAM-MLP-Binary-SGD: Test set: Average loss: 0.0524, Accuracy: 31/40 (78%)\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z5w-7O5jsMi2"
      },
      "source": [
        "## 2.e. Evaluate model"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.plot(train_counter, train_losses, color='blue')\n",
        "plt.plot(test_counter, test_losses, color='red')\n",
        "plt.legend(['Loss Per Epoch -BIGRAM-MLP-Binary-SGD:,', 'Test Loss'], loc='upper right')\n",
        "plt.xlabel('#Training Data Processed')\n",
        "plt.ylabel('Negative Log Likelihood Loss')\n",
        "plt.show()\n",
        "\n",
        "precision, recall, f1_scores = evaluate_preds(predictions, targets)\n",
        "visualize_evaluation(class_to_idx, precision, recall, f1_scores)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 688
        },
        "id": "jb4emYxRCpNP",
        "outputId": "94264f6d-b44d-4050-e692-c466b0813382"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "  Labels/Malwares   Precision     Recall   F1 Score\n",
            "0           Agent    0.000000   0.000000   0.000000\n",
            "1         Allaple  100.000000  30.666667  46.938776\n",
            "2          AutoIt  100.000000  78.000000  87.640449\n",
            "3           Basun   50.000000  90.000000  64.285714\n",
            "4    NothingFound   55.189873  62.285714  58.523490\n",
            "5         Patched    0.000000   0.000000   0.000000\n",
            "6         Swizzor   75.862069  44.000000  55.696203\n",
            "7           Texel   70.727848  81.272727  75.634518\n",
            "8              VB    0.000000   0.000000   0.000000\n",
            "9           Virut   76.800000  48.000000  59.076923\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U_GE1BQjsMi3"
      },
      "source": [
        "## 2.f. Save Model + Data"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "expt_name = 'BIGRAM-MLP-Binary-SGD:'\n",
        "\n",
        "model_out_path = '{}.checkpoint.pth'.format(expt_name)\n",
        "save_model(network, model_out_path)\n",
        "\n",
        "eval_data = {\n",
        "    'epoch': n_epochs,\n",
        "    'train_loss': train_losses,\n",
        "    'test_loss': test_losses,\n",
        "    'train_acc': [],\n",
        "    'test_acc': test_accuracies\n",
        "}\n",
        "eval_out_path = '{}.eval.pickle'.format(expt_name)\n",
        "save_data(eval_data, eval_out_path)"
      ],
      "metadata": {
        "id": "XF0yaHf4C1wE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uLD6dur1sMi3"
      },
      "source": [
        "# 3. Analysis"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1QRKxLsPsMi3"
      },
      "source": [
        "## 3.a. Summary: Main Results\n",
        "\n",
        "We tried different configurations to compute the test loss and accuracy which are summarised as follows:\n",
        "\n",
        "S.no | Model-Name  | Vectorizer | Network | Optimizer | Test Loss | Accuracy  \n",
        "-----|-------------|------------|---------|-----------|----------|----------\n",
        "1    |    Unigram-Model-1  |  Count | MLP | SGD |  0.059 |  75% \n",
        "2    |    Unigram-Model-2  |  Binary | MLP | SGD | 0.059 |  75%\n",
        "3    |    Bigram-Model-1  |  Count | MLP | SGD |  0.051 |  76% \n",
        "4    |    Bigram-Model-2  |  Binary | MLP | SGD | 0.052 |  78%\n",
        "5    |    Trigram-Model-1  |  Count | MLP | SGD |  0.055 |  78% \n",
        "6    |    Trigram-Model-2  |  Binary | MLP | SGD | 0.053 |  79%\n",
        "7    |    Unigram-Model-3  |  Count | MLP | ADAMW |  0.057 |  76% \n",
        "8    |    Unigram-Model-4  |  Binary | MLP | ADAMW | 0.052 |  72%\n",
        "9    |    Bigram-Model-3  |  Count | MLP | ADAMW |  0.047 |  81% \n",
        "10    |    Bigram-Model-4  |  Binary | MLP | ADAMW | 0.049 |  79%\n",
        "11    |    Trigram-Model-3  |  Count | MLP | ADAMW |  0.048 |  80% \n",
        "12    |    Trigram-Model-4  |  Binary | MLP | ADAMW | 0.055 |  78%\n",
        "\n",
        "\n",
        "Due to limitations of computational power, we restricted to using 500 samples for the testing and training purposes and used a 90-10% split for the training and testing respectively (Thus we see a shorter testing loss graph.\n",
        "\n",
        "We created various models by choosing various types of n-grams, vectoriser and \n",
        "moreover, we also played around with various optimisers like Stochastic Gradient Descent(SGD) with momentum and ADAMW which is a newer version of ADAM which is known for improving generalisation loss and test accuracy.\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "collapsed": true,
        "id": "dNKqJAoasMi4"
      },
      "source": [
        "## 3.b. Discussion\n",
        "\n",
        "Enter your final summary here.\n",
        "\n",
        "For instance, you can address:\n",
        "- What was the performance you obtained with the simplest approach?\n",
        "- Which vectorized input representations helped more than the others?\n",
        "- Which malwares are difficult to detect and why?\n",
        "- Which approach do you recommend to perform malware classification?"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "\n",
        "\n",
        "*  The simplest model was Unigram with Binary vectorizer which gave us an accuracy of 75% and 72% with SGD and ADAMW optimizer respectively.\n",
        "*   We used two types of vectorizer, the first vectorizer is of binary type which gives an output of 1/0 based on if a word is present in the input and the the other vectoriser, count type not only just considers the presence but also gives the frequency of occurance. \n",
        "    \n",
        "    Based on the accuracy of various models, it is difficult to say which one performed better as the performance using both the vectorizer is comparable.\n",
        "\n",
        "*   The malwares that are difficult to detect are 'Agent', 'Patched' and 'VB' because in our opinion, they might need higher magnitude of n-grams to make the classification, which in our case unigram, bigrams and trigrams failed to achieve.\n",
        "\n",
        "*   In our opinion, we feel a MLP was the best choice for the malware classification as it doesn't require in-depth knowledge of the database and alongside a good optimizer like SGD+momentum or ADAM or others, the accuracy achieved is high. But while using MLP we also need to set the parameters by careful consideration as they play a major role for getting a good test accuracy.\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "-p1OoREIQIQU"
      }
    }
  ]
}