Q2.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "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": "3.6.10"
    },
    "colab": {
      "name": "Q2 (A2).ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "AIldeHYFipEN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# !pip install pyunpack\n",
        "# !pip install patool"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lJD91Ep72TYz",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f84755a2-74c6-419a-9d7e-1b158b1a0bd8"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s7xRvOOGcsL7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# from pyunpack import Archive\n",
        "# Archive('/content/drive/My Drive/Data_Set_2(Black_and_white_images).rar').extractall('/content/drive/My Drive/Data_Set_2(Black_and_white_images)')"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JTvZrj6Tb9CW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch\n",
        "import torchvision\n",
        "import torchvision.transforms as transforms\n",
        "import tarfile\n",
        "import pandas as pd\n",
        "import os\n",
        "import re\n",
        "from torch.utils.data import Dataset, DataLoader, ConcatDataset, random_split\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import torch.optim as optim\n",
        "from sklearn.metrics import confusion_matrix\n",
        "from sklearn.decomposition import PCA\n",
        "from io import StringIO\n",
        "from sklearn.metrics import accuracy_score, f1_score, precision_score\n",
        "import pickle"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7nIBihjGb9Cm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class DatasetClass(Dataset):\n",
        "    \n",
        "    def __init__(self, folder, filename, label_dict):\n",
        "        \n",
        "        self.data = []\n",
        "        self.filename = filename\n",
        "        tar = tarfile.open(folder + '/' + filename)\n",
        "        for file in tar.getmembers():\n",
        "            f = tar.extractfile(file)\n",
        "            if f != None:\n",
        "                content = pd.read_csv(StringIO(f.read().decode()), sep=' ', header=None).values.ravel()\n",
        "                self.data.append(content)\n",
        "            \n",
        "        self.y = torch.tensor(label_dict[self.filename[:-7]], dtype=torch.long)\n",
        "    \n",
        "    def __getitem__(self, idx):     \n",
        "        \n",
        "        return torch.tensor(self.data[idx], dtype=torch.float), self.y\n",
        "      \n",
        "    def __len__(self):\n",
        "        \n",
        "        return len(self.data)"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "abtj_RwNb9Cw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train_test_loader(directory, label_dict, train_fraction=0.8, num_workers=2):\n",
        "\n",
        "    all_files = list(filter(lambda x: x.endswith('.tar.gz'), os.listdir(directory)))\n",
        "    files = [file for file in all_files if file[:-7] in label_dict.keys()]\n",
        "    \n",
        "    datasets = list(map(lambda x : DatasetClass(directory, x, label_dict), files))\n",
        "    dataset = ConcatDataset(datasets)\n",
        "    N = dataset.cumulative_sizes[-1]\n",
        "    \n",
        "    train_size = int(N*train_fraction)\n",
        "    test_size = N - train_size\n",
        "\n",
        "    train_data, test_data = torch.utils.data.random_split(dataset, [train_size, test_size])\n",
        "\n",
        "    trainloader = DataLoader(train_data, batch_size=32, shuffle=True, num_workers=num_workers)\n",
        "    testloader = DataLoader(test_data, batch_size=32, shuffle=True, num_workers=num_workers)\n",
        "    \n",
        "    return trainloader, testloader, train_size, test_size"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tRYvvEcXb9C5",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "label_dict = {\n",
        "    'tallbuilding': 0,\n",
        "    'mountain': 1,\n",
        "    'highway': 2,\n",
        "    'coast': 3, \n",
        "    'opencountry': 4}\n",
        "\n",
        "trainloader, testloader, train_size, test_size = train_test_loader('/content/drive/My Drive/Data_Set_1(Colored_Images)', label_dict, train_fraction=0.8, num_workers=0)"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1hHs6vC6b9C-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class AutoEncoder(nn.Module):\n",
        "    \n",
        "    def __init__(self, n_features, h_layer_sizes):\n",
        "        super(AutoEncoder, self).__init__()\n",
        "        \n",
        "        self.fc1 = nn.Linear(n_features, h_layer_sizes[0])\n",
        "        self.fc2 = nn.Linear(h_layer_sizes[0], h_layer_sizes[1])\n",
        "        self.fc3 = nn.Linear(h_layer_sizes[1], h_layer_sizes[2])\n",
        "        self.out = nn.Linear(h_layer_sizes[2], n_features)\n",
        "        \n",
        "    def forward(self, x):\n",
        "        \n",
        "        x = torch.tanh(self.fc1(x)) # Hidden Layer 1 (Tanh)\n",
        "        x = self.fc2(x)    # Hidden Layer 2 (Linear)\n",
        "        x = torch.tanh(self.fc3(x)) # Hidden Layer 3 (Tanh)\n",
        "        x = self.out(x) # Output Layer (Linear)\n",
        "        \n",
        "        return x\n",
        "    \n",
        "    def get_z(self, x):\n",
        "        \n",
        "        z = torch.tanh(self.fc1(x))\n",
        "        z = self.fc2(z)\n",
        "        \n",
        "        return z"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0e_pNIyEb9DE",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "898257e8-39fd-4b56-e85f-422e4867c764"
      },
      "source": [
        "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
        "print(device)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "cuda:0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZSWJgWEucK0r",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "X_train = []\n",
        "for X, y in trainloader:\n",
        "  X_train.extend(X.numpy())\n",
        "X_train = np.array(X_train)"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BEICm-tqcEBv",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "7a73dc8c-37d2-4f55-d172-11b232c447c8"
      },
      "source": [
        "pca1 = PCA(n_components=X_train.shape[1])\n",
        "pca1.fit(X_train)\n",
        "\n",
        "print('97% Variance Explained:', np.where(np.cumsum(pca1.explained_variance_ratio_)>=0.97)[0][0]+1)\n",
        "explained_var = np.cumsum(pca1.explained_variance_ratio_)\n",
        "plt.plot(explained_var)\n",
        "plt.show()"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "97% Variance Explained: 228\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ru64o3f2b9DO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "criterion = nn.MSELoss()\n",
        "\n",
        "ae1 = AutoEncoder(828, [400, 200, 400])\n",
        "ae1 = ae1.to(device)\n",
        "optimizer1 = optim.SGD(ae1.parameters(), lr=0.0001, momentum=0.9)"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "trgn2rUeb9DY",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "6a79a976-100c-4065-f9b6-8e2ca67b9bb2"
      },
      "source": [
        "old_loss = np.inf\n",
        "\n",
        "max_epoch = 500\n",
        "losses = []\n",
        "for epoch in range(max_epoch):\n",
        "\n",
        "    running_loss = 0.0\n",
        "    len_ = 0\n",
        "    \n",
        "    for data in trainloader:\n",
        "        \n",
        "        X, _ = data[0].to(device), data[1].to(device)\n",
        "        \n",
        "        optimizer1.zero_grad()\n",
        "        \n",
        "        # Reconstructed Representation of X (forward)\n",
        "        X_hat = ae1(X)\n",
        "        \n",
        "        # Calculate Loss (MSE)\n",
        "        loss = criterion(X_hat, X)\n",
        "        \n",
        "        # Backpropagation\n",
        "        loss.backward()\n",
        "        \n",
        "        # Update Parameters\n",
        "        optimizer1.step()\n",
        "        \n",
        "        running_loss += loss.item()*len(X)/train_size\n",
        "    \n",
        "    print('Epoch', epoch+1, ': Loss =', running_loss)\n",
        "    losses.append(running_loss)\n",
        "    \n",
        "    if (abs(running_loss-old_loss)/running_loss < 1e-4) and running_loss<0.201:\n",
        "        print('Converged')\n",
        "        break\n",
        "    \n",
        "    old_loss = running_loss\n",
        "\n",
        "print('Finished Training')\n",
        "plt.plot(losses)\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Iter Number')\n",
        "plt.title('Convergence monitor plot')\n",
        "plt.show()"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1 : Loss = 1.9403873682022095\n",
            "Epoch 2 : Loss = 1.9177996868436988\n",
            "Epoch 3 : Loss = 1.895418722521175\n",
            "Epoch 4 : Loss = 1.8738206733356826\n",
            "Epoch 5 : Loss = 1.852434567429802\n",
            "Epoch 6 : Loss = 1.8307957622137936\n",
            "Epoch 7 : Loss = 1.80853272839026\n",
            "Epoch 8 : Loss = 1.7852746194059197\n",
            "Epoch 9 : Loss = 1.7608684030446136\n",
            "Epoch 10 : Loss = 1.7350585569034929\n",
            "Epoch 11 : Loss = 1.7076921544291765\n",
            "Epoch 12 : Loss = 1.6787846305153584\n",
            "Epoch 13 : Loss = 1.6481998589905824\n",
            "Epoch 14 : Loss = 1.6159914217211981\n",
            "Epoch 15 : Loss = 1.5821706273339011\n",
            "Epoch 16 : Loss = 1.546823943203146\n",
            "Epoch 17 : Loss = 1.5100137103687639\n",
            "Epoch 18 : Loss = 1.471907621080225\n",
            "Epoch 19 : Loss = 1.4326800378886135\n",
            "Epoch 20 : Loss = 1.3924544887109234\n",
            "Epoch 21 : Loss = 1.351450622081756\n",
            "Epoch 22 : Loss = 1.3099071031266991\n",
            "Epoch 23 : Loss = 1.2679279256950726\n",
            "Epoch 24 : Loss = 1.2258631966330786\n",
            "Epoch 25 : Loss = 1.1837784024802123\n",
            "Epoch 26 : Loss = 1.141958174380389\n",
            "Epoch 27 : Loss = 1.1005016429857777\n",
            "Epoch 28 : Loss = 1.0596684434197166\n",
            "Epoch 29 : Loss = 1.0195417065512054\n",
            "Epoch 30 : Loss = 0.9801955250176514\n",
            "Epoch 31 : Loss = 0.9418841857801784\n",
            "Epoch 32 : Loss = 0.9046229760755191\n",
            "Epoch 33 : Loss = 0.8684398748657921\n",
            "Epoch 34 : Loss = 0.8334655368869952\n",
            "Epoch 35 : Loss = 0.7997106042775239\n",
            "Epoch 36 : Loss = 0.7672609646211971\n",
            "Epoch 37 : Loss = 0.7360542606223714\n",
            "Epoch 38 : Loss = 0.7061626802791248\n",
            "Epoch 39 : Loss = 0.6775536184961148\n",
            "Epoch 40 : Loss = 0.6502630629322746\n",
            "Epoch 41 : Loss = 0.6241977702487598\n",
            "Epoch 42 : Loss = 0.5993999391794205\n",
            "Epoch 43 : Loss = 0.5758101466027172\n",
            "Epoch 44 : Loss = 0.5534449057145553\n",
            "Epoch 45 : Loss = 0.5321845507079903\n",
            "Epoch 46 : Loss = 0.5120928612622349\n",
            "Epoch 47 : Loss = 0.4930482378060167\n",
            "Epoch 48 : Loss = 0.475045603784648\n",
            "Epoch 49 : Loss = 0.4580514742569489\n",
            "Epoch 50 : Loss = 0.442002429880879\n",
            "Epoch 51 : Loss = 0.42689860679886565\n",
            "Epoch 52 : Loss = 0.4126270806247538\n",
            "Epoch 53 : Loss = 0.3992198001254688\n",
            "Epoch 54 : Loss = 0.38657588850368146\n",
            "Epoch 55 : Loss = 0.374697316099297\n",
            "Epoch 56 : Loss = 0.36353820359165007\n",
            "Epoch 57 : Loss = 0.3530428226698528\n",
            "Epoch 58 : Loss = 0.3432156294584275\n",
            "Epoch 59 : Loss = 0.33395473862236197\n",
            "Epoch 60 : Loss = 0.32529474253004254\n",
            "Epoch 61 : Loss = 0.3171620335091244\n",
            "Epoch 62 : Loss = 0.3095403286543759\n",
            "Epoch 63 : Loss = 0.3024014796723018\n",
            "Epoch 64 : Loss = 0.2957233034751633\n",
            "Epoch 65 : Loss = 0.28945017199624673\n",
            "Epoch 66 : Loss = 0.2835818874565037\n",
            "Epoch 67 : Loss = 0.27810741012746637\n",
            "Epoch 68 : Loss = 0.27296770838173956\n",
            "Epoch 69 : Loss = 0.26816125044768513\n",
            "Epoch 70 : Loss = 0.263673987578262\n",
            "Epoch 71 : Loss = 0.2594755434177139\n",
            "Epoch 72 : Loss = 0.2555447976020249\n",
            "Epoch 73 : Loss = 0.25187959522008896\n",
            "Epoch 74 : Loss = 0.24843869324434883\n",
            "Epoch 75 : Loss = 0.24524416469714863\n",
            "Epoch 76 : Loss = 0.24225612458857623\n",
            "Epoch 77 : Loss = 0.2394454472444274\n",
            "Epoch 78 : Loss = 0.23684154721823603\n",
            "Epoch 79 : Loss = 0.234394088726152\n",
            "Epoch 80 : Loss = 0.23212075978517532\n",
            "Epoch 81 : Loss = 0.22999842058528552\n",
            "Epoch 82 : Loss = 0.227997692132538\n",
            "Epoch 83 : Loss = 0.22615469450300388\n",
            "Epoch 84 : Loss = 0.22442080215974286\n",
            "Epoch 85 : Loss = 0.22280028733340174\n",
            "Epoch 86 : Loss = 0.22129705717617815\n",
            "Epoch 87 : Loss = 0.21988382969390263\n",
            "Epoch 88 : Loss = 0.21857039291750296\n",
            "Epoch 89 : Loss = 0.21734232658689673\n",
            "Epoch 90 : Loss = 0.21619987657124348\n",
            "Epoch 91 : Loss = 0.2151335521855138\n",
            "Epoch 92 : Loss = 0.21412954242392013\n",
            "Epoch 93 : Loss = 0.21320457282391456\n",
            "Epoch 94 : Loss = 0.21233283051035612\n",
            "Epoch 95 : Loss = 0.21152493865652514\n",
            "Epoch 96 : Loss = 0.21076574142683632\n",
            "Epoch 97 : Loss = 0.21006038954312153\n",
            "Epoch 98 : Loss = 0.20940075476061218\n",
            "Epoch 99 : Loss = 0.20878857204859913\n",
            "Epoch 100 : Loss = 0.2082115137441592\n",
            "Epoch 101 : Loss = 0.20767702602527355\n",
            "Epoch 102 : Loss = 0.20717823234471414\n",
            "Epoch 103 : Loss = 0.20671057497913192\n",
            "Epoch 104 : Loss = 0.20627473938194188\n",
            "Epoch 105 : Loss = 0.20587153800509195\n",
            "Epoch 106 : Loss = 0.20548748631368982\n",
            "Epoch 107 : Loss = 0.2051357332278382\n",
            "Epoch 108 : Loss = 0.20480416884476488\n",
            "Epoch 109 : Loss = 0.2044952342456037\n",
            "Epoch 110 : Loss = 0.20420722595670007\n",
            "Epoch 111 : Loss = 0.20393812927332786\n",
            "Epoch 112 : Loss = 0.2036855149675499\n",
            "Epoch 113 : Loss = 0.2034514299170537\n",
            "Epoch 114 : Loss = 0.2032322724434463\n",
            "Epoch 115 : Loss = 0.2030266655439681\n",
            "Epoch 116 : Loss = 0.202834602445364\n",
            "Epoch 117 : Loss = 0.2026569182899865\n",
            "Epoch 118 : Loss = 0.20248939333991572\n",
            "Epoch 119 : Loss = 0.2023320367390459\n",
            "Epoch 120 : Loss = 0.20218831266869197\n",
            "Epoch 121 : Loss = 0.20204959809780115\n",
            "Epoch 122 : Loss = 0.20192266695878724\n",
            "Epoch 123 : Loss = 0.2018030153756792\n",
            "Epoch 124 : Loss = 0.20169179751114408\n",
            "Epoch 125 : Loss = 0.20158656551079313\n",
            "Epoch 126 : Loss = 0.20148940011858943\n",
            "Epoch 127 : Loss = 0.20139750194820488\n",
            "Epoch 128 : Loss = 0.2013116414574059\n",
            "Epoch 129 : Loss = 0.20123217356475917\n",
            "Epoch 130 : Loss = 0.20115601339123465\n",
            "Epoch 131 : Loss = 0.20108589699322527\n",
            "Epoch 132 : Loss = 0.2010211697356267\n",
            "Epoch 133 : Loss = 0.20095817473801697\n",
            "Epoch 134 : Loss = 0.20090104334733705\n",
            "Epoch 135 : Loss = 0.20084596526893703\n",
            "Epoch 136 : Loss = 0.20079427991401066\n",
            "Epoch 137 : Loss = 0.200746741484512\n",
            "Epoch 138 : Loss = 0.20070205764337026\n",
            "Epoch 139 : Loss = 0.20065936344591048\n",
            "Epoch 140 : Loss = 0.2006199928847226\n",
            "Epoch 141 : Loss = 0.20058194412426514\n",
            "Epoch 142 : Loss = 0.20054705407131804\n",
            "Epoch 143 : Loss = 0.20051405985247006\n",
            "Epoch 144 : Loss = 0.2004828510636633\n",
            "Epoch 145 : Loss = 0.20045400986617262\n",
            "Epoch 146 : Loss = 0.20042536949569534\n",
            "Epoch 147 : Loss = 0.200399728322571\n",
            "Epoch 148 : Loss = 0.20037466389211747\n",
            "Epoch 149 : Loss = 0.2003519470718774\n",
            "Epoch 150 : Loss = 0.20032945918765938\n",
            "Epoch 151 : Loss = 0.2003080861812288\n",
            "Epoch 152 : Loss = 0.2002889879725196\n",
            "Converged\n",
            "Finished Training\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IzLhTvpq1zft",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "b03a8b56-64a4-480c-9851-a420958280d0"
      },
      "source": [
        "test_loss = 0.0\n",
        "\n",
        "for data in testloader:\n",
        "        \n",
        "    X, _ = data[0].to(device), data[1].to(device)\n",
        "    with torch.no_grad():\n",
        "      X_pred = ae1(X)\n",
        "      loss = criterion(X_pred, X)\n",
        "\n",
        "    test_loss += loss.item()*len(X)/test_size\n",
        "\n",
        "test_loss = test_loss\n",
        "print(test_loss)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.2036638016050512\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kQpvQyBK5CnJ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "Z1_array = []\n",
        "\n",
        "for data in trainloader:\n",
        "        \n",
        "    X, _ = data[0].to(device), data[1].to(device)\n",
        "    with torch.no_grad():\n",
        "      Z1 = ae1.get_z(X)\n",
        "\n",
        "    if len(Z1_array) == 0:\n",
        "        Z1_array = Z1.cpu().numpy()\n",
        "    else:\n",
        "        Z1_array = np.append(Z1_array, Z1.cpu().numpy(), axis=0)"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TA2o4c0KivOE",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "f01cd294-28c4-44ad-96e2-921e89ab8f44"
      },
      "source": [
        "pca2 = PCA(n_components=Z1_array.shape[1])\n",
        "pca2.fit(Z1_array)\n",
        "print('97% Variance Explained:', np.where(np.cumsum(pca2.explained_variance_ratio_)>=0.97)[0][0]+1)\n",
        "\n",
        "explained_var = np.cumsum(pca2.explained_variance_ratio_)\n",
        "plt.plot(explained_var)\n",
        "plt.show()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "97% Variance Explained: 76\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pwcvgUCVb9Df",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "ae2 = AutoEncoder(200, [150, 75, 150])\n",
        "optimizer2 = optim.SGD(ae2.parameters(), lr=0.0001, momentum=0.9)\n",
        "ae2 = ae2.to(device)"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J144k8cub9Dn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "98822164-b4b8-4e57-8edc-a450a4d1a5dd"
      },
      "source": [
        "old_loss = np.inf\n",
        "losses = []\n",
        "max_epoch = 500\n",
        "\n",
        "for epoch in range(max_epoch):\n",
        "\n",
        "    running_loss = 0.0\n",
        "    \n",
        "    for data in trainloader:\n",
        "        \n",
        "        X, _ = data[0].to(device), data[1].to(device)\n",
        "        \n",
        "        with torch.no_grad():\n",
        "            Z = ae1.get_z(X)\n",
        "        \n",
        "        optimizer2.zero_grad()\n",
        "        \n",
        "        # Reconstructed Representation of X (forward)\n",
        "        Z_hat = ae2(Z)\n",
        "        \n",
        "        # Calculate Loss (MSE)\n",
        "        loss = criterion(Z_hat, Z)\n",
        "        \n",
        "        # Backpropagation\n",
        "        loss.backward()\n",
        "        \n",
        "        # Update Parameters\n",
        "        optimizer2.step()\n",
        "        \n",
        "        running_loss += loss.item()*len(X)/train_size\n",
        "    \n",
        "    losses.append(running_loss)\n",
        "    print('Epoch', epoch+1, ': Loss =', running_loss)\n",
        "    \n",
        "    if abs(running_loss-old_loss)/running_loss < 0.01 and running_loss<9e-3:\n",
        "        print('Converged')\n",
        "        break\n",
        "        \n",
        "    old_loss = running_loss\n",
        "\n",
        "print('Finished Training')\n",
        "plt.plot(losses)\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Iter Number')\n",
        "plt.title('Convergence monitor plot')\n",
        "plt.show()"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1 : Loss = 2.486224949359894\n",
            "Epoch 2 : Loss = 2.4323756965723917\n",
            "Epoch 3 : Loss = 2.373781751502644\n",
            "Epoch 4 : Loss = 2.3116262500936333\n",
            "Epoch 5 : Loss = 2.245121278546073\n",
            "Epoch 6 : Loss = 2.1738502708348357\n",
            "Epoch 7 : Loss = 2.0976461822336376\n",
            "Epoch 8 : Loss = 2.016669798981059\n",
            "Epoch 9 : Loss = 1.9312594614245666\n",
            "Epoch 10 : Loss = 1.8419459895654162\n",
            "Epoch 11 : Loss = 1.749526514248415\n",
            "Epoch 12 : Loss = 1.654785167087208\n",
            "Epoch 13 : Loss = 1.5587453110651532\n",
            "Epoch 14 : Loss = 1.4622688022526829\n",
            "Epoch 15 : Loss = 1.3663457550785756\n",
            "Epoch 16 : Loss = 1.2718696783889425\n",
            "Epoch 17 : Loss = 1.1796076351946048\n",
            "Epoch 18 : Loss = 1.0903383628888565\n",
            "Epoch 19 : Loss = 1.0045466599139297\n",
            "Epoch 20 : Loss = 0.9227807318622416\n",
            "Epoch 21 : Loss = 0.845357372002168\n",
            "Epoch 22 : Loss = 0.7724495950070294\n",
            "Epoch 23 : Loss = 0.7042089605873281\n",
            "Epoch 24 : Loss = 0.6406712599776008\n",
            "Epoch 25 : Loss = 0.58179073035717\n",
            "Epoch 26 : Loss = 0.5274055498567493\n",
            "Epoch 27 : Loss = 0.47739270058545186\n",
            "Epoch 28 : Loss = 0.43153201314536005\n",
            "Epoch 29 : Loss = 0.3896257775751028\n",
            "Epoch 30 : Loss = 0.35142254490744\n",
            "Epoch 31 : Loss = 0.3166672465476123\n",
            "Epoch 32 : Loss = 0.28515197065743536\n",
            "Epoch 33 : Loss = 0.2565759376368739\n",
            "Epoch 34 : Loss = 0.23075860027562492\n",
            "Epoch 35 : Loss = 0.20743718675591724\n",
            "Epoch 36 : Loss = 0.18643583458932964\n",
            "Epoch 37 : Loss = 0.1675110225650397\n",
            "Epoch 38 : Loss = 0.15049288049340245\n",
            "Epoch 39 : Loss = 0.13519874655387623\n",
            "Epoch 40 : Loss = 0.12147618762471458\n",
            "Epoch 41 : Loss = 0.10916881720450791\n",
            "Epoch 42 : Loss = 0.09813730350949547\n",
            "Epoch 43 : Loss = 0.08826215903867378\n",
            "Epoch 44 : Loss = 0.07941981997679581\n",
            "Epoch 45 : Loss = 0.07150927694006401\n",
            "Epoch 46 : Loss = 0.0644403300671415\n",
            "Epoch 47 : Loss = 0.05811665249480443\n",
            "Epoch 48 : Loss = 0.05246955020861192\n",
            "Epoch 49 : Loss = 0.04743157377974554\n",
            "Epoch 50 : Loss = 0.04292764616283503\n",
            "Epoch 51 : Loss = 0.03891290673478083\n",
            "Epoch 52 : Loss = 0.03532438318837772\n",
            "Epoch 53 : Loss = 0.03212711761112917\n",
            "Epoch 54 : Loss = 0.02927715407515114\n",
            "Epoch 55 : Loss = 0.0267289852808145\n",
            "Epoch 56 : Loss = 0.024464454078538856\n",
            "Epoch 57 : Loss = 0.02243640985001217\n",
            "Epoch 58 : Loss = 0.020637546962296423\n",
            "Epoch 59 : Loss = 0.019028611354191198\n",
            "Epoch 60 : Loss = 0.01759590545076538\n",
            "Epoch 61 : Loss = 0.016320643290369346\n",
            "Epoch 62 : Loss = 0.015178914330053056\n",
            "Epoch 63 : Loss = 0.014166880504820834\n",
            "Epoch 64 : Loss = 0.01326080583120612\n",
            "Epoch 65 : Loss = 0.012454610635441815\n",
            "Epoch 66 : Loss = 0.011738649040291255\n",
            "Epoch 67 : Loss = 0.011096721028231765\n",
            "Epoch 68 : Loss = 0.010525327916680415\n",
            "Epoch 69 : Loss = 0.0100176527842202\n",
            "Epoch 70 : Loss = 0.00956356548704207\n",
            "Epoch 71 : Loss = 0.009160171739164401\n",
            "Epoch 72 : Loss = 0.008799350354820486\n",
            "Epoch 73 : Loss = 0.008478724461218171\n",
            "Epoch 74 : Loss = 0.008191620613532988\n",
            "Epoch 75 : Loss = 0.00793694007337432\n",
            "Epoch 76 : Loss = 0.007709073982286184\n",
            "Epoch 77 : Loss = 0.007506890804506839\n",
            "Epoch 78 : Loss = 0.007325128006579523\n",
            "Epoch 79 : Loss = 0.007163668273609471\n",
            "Epoch 80 : Loss = 0.007019748974760823\n",
            "Epoch 81 : Loss = 0.006891361543570053\n",
            "Epoch 82 : Loss = 0.006778071137060497\n",
            "Epoch 83 : Loss = 0.0066743929320099695\n",
            "Epoch 84 : Loss = 0.006583225172521039\n",
            "Epoch 85 : Loss = 0.006502411878583106\n",
            "Epoch 86 : Loss = 0.0064296322260898615\n",
            "Epoch 87 : Loss = 0.006364604997956617\n",
            "Epoch 88 : Loss = 0.006307027577845888\n",
            "Converged\n",
            "Finished Training\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6VHpK73lgdGV",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f6e726d9-c82c-413e-ec4f-c7222a0d7cf9"
      },
      "source": [
        "test_loss = 0.0\n",
        "\n",
        "for data in testloader:\n",
        "        \n",
        "    X, _ = data[0].to(device), data[1].to(device)\n",
        "    with torch.no_grad():\n",
        "      Z = ae1.get_z(X)\n",
        "      Z_pred = ae2(Z)\n",
        "      loss = criterion(Z_pred, Z)\n",
        "\n",
        "    test_loss += loss.item()*len(X)/test_size\n",
        "\n",
        "test_loss = test_loss\n",
        "print(test_loss)"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.006431612948125059\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ie7JQXLshlWK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "Z2_array = []\n",
        "\n",
        "for data in trainloader:\n",
        "        \n",
        "    X, _ = data[0].to(device), data[1].to(device)\n",
        "    with torch.no_grad():\n",
        "      Z1 = ae1.get_z(X)\n",
        "      Z2 = ae2.get_z(Z1)\n",
        "\n",
        "    if len(Z2_array) == 0:\n",
        "        Z2_array = Z2.cpu().numpy()\n",
        "    else:\n",
        "        Z2_array = np.append(Z2_array, Z2.cpu().numpy(), axis=0)"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ySCjdA11h5nq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "e313f3be-d75d-451f-ae7f-722714cc2287"
      },
      "source": [
        "pca3 = PCA(n_components=Z2_array.shape[1])\n",
        "pca3.fit(Z2_array)\n",
        "print('97% Variance Explained:', np.where(np.cumsum(pca3.explained_variance_ratio_)>=0.97)[0][0]+1)\n",
        "\n",
        "explained_var = np.cumsum(pca3.explained_variance_ratio_)\n",
        "plt.plot(explained_var)\n",
        "plt.show()"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "97% Variance Explained: 11\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TVEyxUUBb9Dt",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "ae3 = AutoEncoder(75, [50, 15, 50])\n",
        "optimizer3 = optim.SGD(ae3.parameters(), lr=0.0001, momentum=0.9)\n",
        "ae3 = ae3.to(device)"
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wV0pjnuNb9Dy",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "ec595872-6468-46cb-cd7f-a3fe3f94ce7c"
      },
      "source": [
        "old_loss = np.inf\n",
        "\n",
        "max_epoch = 500\n",
        "losses = []\n",
        "\n",
        "for epoch in range(max_epoch):\n",
        "\n",
        "    running_loss = 0.0\n",
        "    for data in trainloader:\n",
        "        \n",
        "        X, _ = data[0].to(device), data[1].to(device)\n",
        "        \n",
        "        with torch.no_grad():\n",
        "            Z1 = ae1.get_z(X)\n",
        "            Z = ae2.get_z(Z1)\n",
        "        \n",
        "        optimizer3.zero_grad()\n",
        "        \n",
        "        # Reconstructed Representation of Z (forward)\n",
        "        Z_hat = ae3(Z)\n",
        "        \n",
        "        # Calculate Loss (MSE)\n",
        "        loss = criterion(Z_hat, Z)\n",
        "        \n",
        "        # Backpropagation\n",
        "        loss.backward()\n",
        "        \n",
        "        # Update Parameters\n",
        "        optimizer3.step()\n",
        "        \n",
        "        running_loss += loss.item()*len(X)/train_size\n",
        "    \n",
        "    print('Epoch', epoch+1, ': Loss =', running_loss)\n",
        "    losses.append(running_loss)\n",
        "\n",
        "    if abs(running_loss-old_loss)/running_loss < 0.05 and running_loss<1e-3:\n",
        "        print('Converged')\n",
        "        break\n",
        "        \n",
        "    old_loss = running_loss\n",
        "\n",
        "print('Finished Training')\n",
        "plt.plot(losses)\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Iter Number')\n",
        "plt.title('Convergence monitor plot')\n",
        "plt.show()"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1 : Loss = 2.7355596314777033\n",
            "Epoch 2 : Loss = 2.673307229172099\n",
            "Epoch 3 : Loss = 2.609427863901311\n",
            "Epoch 4 : Loss = 2.5444942658597762\n",
            "Epoch 5 : Loss = 2.476547067815607\n",
            "Epoch 6 : Loss = 2.404238722541115\n",
            "Epoch 7 : Loss = 2.3268609155308115\n",
            "Epoch 8 : Loss = 2.2442985718900506\n",
            "Epoch 9 : Loss = 2.156820232217962\n",
            "Epoch 10 : Loss = 2.064958323131908\n",
            "Epoch 11 : Loss = 1.969396255233071\n",
            "Epoch 12 : Loss = 1.8710036955096512\n",
            "Epoch 13 : Loss = 1.770762828263369\n",
            "Epoch 14 : Loss = 1.6697434430772609\n",
            "Epoch 15 : Loss = 1.5689754296432845\n",
            "Epoch 16 : Loss = 1.4694481762972746\n",
            "Epoch 17 : Loss = 1.3720183670520778\n",
            "Epoch 18 : Loss = 1.277424877340143\n",
            "Epoch 19 : Loss = 1.1862492994828662\n",
            "Epoch 20 : Loss = 1.0989422391761436\n",
            "Epoch 21 : Loss = 1.0158225758509203\n",
            "Epoch 22 : Loss = 0.9370876293290746\n",
            "Epoch 23 : Loss = 0.862868684259328\n",
            "Epoch 24 : Loss = 0.7931782589717342\n",
            "Epoch 25 : Loss = 0.7279703888026152\n",
            "Epoch 26 : Loss = 0.6671892987056213\n",
            "Epoch 27 : Loss = 0.6106700165705246\n",
            "Epoch 28 : Loss = 0.5582855966958131\n",
            "Epoch 29 : Loss = 0.5098266540603202\n",
            "Epoch 30 : Loss = 0.46511456438086246\n",
            "Epoch 31 : Loss = 0.4239365072412924\n",
            "Epoch 32 : Loss = 0.38608640906485636\n",
            "Epoch 33 : Loss = 0.35134949399666354\n",
            "Epoch 34 : Loss = 0.31951316446065897\n",
            "Epoch 35 : Loss = 0.2903864356604489\n",
            "Epoch 36 : Loss = 0.26376652243462473\n",
            "Epoch 37 : Loss = 0.23946322331374348\n",
            "Epoch 38 : Loss = 0.21729906004938213\n",
            "Epoch 39 : Loss = 0.1971003419973633\n",
            "Epoch 40 : Loss = 0.17872051081874155\n",
            "Epoch 41 : Loss = 0.16199569641189143\n",
            "Epoch 42 : Loss = 0.14679386297410188\n",
            "Epoch 43 : Loss = 0.1329820281402631\n",
            "Epoch 44 : Loss = 0.1204442253166979\n",
            "Epoch 45 : Loss = 0.10906420140103859\n",
            "Epoch 46 : Loss = 0.09874328107319094\n",
            "Epoch 47 : Loss = 0.0893848751756278\n",
            "Epoch 48 : Loss = 0.0809028929268772\n",
            "Epoch 49 : Loss = 0.0732181163674051\n",
            "Epoch 50 : Loss = 0.0662582880732688\n",
            "Epoch 51 : Loss = 0.05995584702627227\n",
            "Epoch 52 : Loss = 0.05424946114759553\n",
            "Epoch 53 : Loss = 0.049086577089672734\n",
            "Epoch 54 : Loss = 0.044412943619218735\n",
            "Epoch 55 : Loss = 0.04018546471541578\n",
            "Epoch 56 : Loss = 0.03636124645444479\n",
            "Epoch 57 : Loss = 0.03290320691567931\n",
            "Epoch 58 : Loss = 0.029774188995361328\n",
            "Epoch 59 : Loss = 0.026946449017321523\n",
            "Epoch 60 : Loss = 0.024387874259528788\n",
            "Epoch 61 : Loss = 0.022077078507705166\n",
            "Epoch 62 : Loss = 0.019985168113965876\n",
            "Epoch 63 : Loss = 0.018096475057642565\n",
            "Epoch 64 : Loss = 0.0163888363590972\n",
            "Epoch 65 : Loss = 0.014844362877986645\n",
            "Epoch 66 : Loss = 0.0134494008962065\n",
            "Epoch 67 : Loss = 0.012187470639632509\n",
            "Epoch 68 : Loss = 0.011047956131568011\n",
            "Epoch 69 : Loss = 0.010017219020731069\n",
            "Epoch 70 : Loss = 0.009086346296085552\n",
            "Epoch 71 : Loss = 0.008244164330376821\n",
            "Epoch 72 : Loss = 0.0074839836452156305\n",
            "Epoch 73 : Loss = 0.006796039651486683\n",
            "Epoch 74 : Loss = 0.006174638444049793\n",
            "Epoch 75 : Loss = 0.005613155969926582\n",
            "Epoch 76 : Loss = 0.005105094867758453\n",
            "Epoch 77 : Loss = 0.004646384898065167\n",
            "Epoch 78 : Loss = 0.004231646582907573\n",
            "Epoch 79 : Loss = 0.0038564056105149735\n",
            "Epoch 80 : Loss = 0.003517462035895071\n",
            "Epoch 81 : Loss = 0.0032111574334769775\n",
            "Epoch 82 : Loss = 0.0029342027899639843\n",
            "Epoch 83 : Loss = 0.0026834276557730677\n",
            "Epoch 84 : Loss = 0.002457293171160431\n",
            "Epoch 85 : Loss = 0.00225234269534915\n",
            "Epoch 86 : Loss = 0.002067295252345502\n",
            "Epoch 87 : Loss = 0.001899965742433613\n",
            "Epoch 88 : Loss = 0.0017483563146511603\n",
            "Epoch 89 : Loss = 0.0016115790984424004\n",
            "Epoch 90 : Loss = 0.0014879462991798807\n",
            "Epoch 91 : Loss = 0.001375768129946664\n",
            "Epoch 92 : Loss = 0.0012744832572273233\n",
            "Epoch 93 : Loss = 0.0011830438134810797\n",
            "Epoch 94 : Loss = 0.00110007396010174\n",
            "Epoch 95 : Loss = 0.0010251361553938216\n",
            "Epoch 96 : Loss = 0.0009572711966360328\n",
            "Epoch 97 : Loss = 0.0008959435085258023\n",
            "Epoch 98 : Loss = 0.0008404689698099073\n",
            "Epoch 99 : Loss = 0.0007903453978625208\n",
            "Epoch 100 : Loss = 0.0007447736014993015\n",
            "Epoch 101 : Loss = 0.0007037563865940849\n",
            "Epoch 102 : Loss = 0.0006664970675377517\n",
            "Epoch 103 : Loss = 0.0006328524759737775\n",
            "Epoch 104 : Loss = 0.0006023367505721662\n",
            "Epoch 105 : Loss = 0.0005748306119295937\n",
            "Converged\n",
            "Finished Training\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "el7d7FcLMTtr",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "6993ca81-bf5a-4fe7-b36f-8de8dcb604a7"
      },
      "source": [
        "test_loss = 0.0\n",
        "\n",
        "for data in testloader:\n",
        "        \n",
        "    X, _ = data[0].to(device), data[1].to(device)\n",
        "    with torch.no_grad():\n",
        "      Z1 = ae1.get_z(X)\n",
        "      Z = ae2.get_z(Z1)\n",
        "      Z_pred = ae3(Z)\n",
        "      loss = criterion(Z_pred, Z)\n",
        "\n",
        "    test_loss += loss.item()*len(X)/test_size\n",
        "\n",
        "test_loss = test_loss\n",
        "print(test_loss)"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.0005464660141363063\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "asYo4AYBb9EB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class FinalNet(nn.Module):\n",
        "    \n",
        "    def __init__(self, input_size, hidden_sizes, num_classes):\n",
        "        super(FinalNet, self).__init__()\n",
        "        \n",
        "        self.fc1 = nn.Linear(input_size, hidden_sizes[0])\n",
        "        self.fc2 = nn.Linear(hidden_sizes[0], hidden_sizes[1])\n",
        "        self.fc3 = nn.Linear(hidden_sizes[1], hidden_sizes[2])\n",
        "        self.fc4 = nn.Linear(hidden_sizes[2], hidden_sizes[3])\n",
        "        self.out = nn.Linear(hidden_sizes[3], num_classes)\n",
        "    \n",
        "    def forward(self, x):\n",
        "        \n",
        "        x = torch.tanh(self.fc1(x))\n",
        "        x = torch.tanh(self.fc2(x))\n",
        "        x = torch.tanh(self.fc3(x))\n",
        "        x = torch.tanh(self.fc4(x))\n",
        "        x = self.out(x)\n",
        "        \n",
        "        return x\n",
        "    \n",
        "    def predict(self, X):\n",
        "        \n",
        "        with torch.no_grad():\n",
        "            y_score = self.forward(X)\n",
        "            y_pred = torch.argmax(y_score, axis=1)\n",
        "            \n",
        "        return y_pred\n",
        "            \n",
        "    \n",
        "classifier = FinalNet(828, [400, 150, 50, 15], 5)"
      ],
      "execution_count": 25,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "azVOLzPcb9EQ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "ae1_params = list(ae1.parameters())\n",
        "ae2_params = list(ae2.parameters())\n",
        "ae3_params = list(ae3.parameters())\n",
        "pickle.dump((ae1_params, ae2_params, ae3_params), open('/content/drive/My Drive/Ass1Q2_init.sav', 'wb'))\n",
        "\n",
        "# (ae1_params, ae2_params, ae3_params) = pickle.load(open('/content/drive/My Drive/Ass1Q2_init.sav', 'rb'))\n",
        "\n",
        "with torch.no_grad():\n",
        "    \n",
        "    classifier.fc1.weight.data = nn.Parameter(ae1_params[0])\n",
        "    classifier.fc1.bias.data = nn.Parameter(ae1_params[1])\n",
        "    \n",
        "    classifier.fc2.weight = nn.Parameter(torch.matmul(ae2_params[0], ae1_params[2]))\n",
        "    classifier.fc2.bias = nn.Parameter(torch.matmul(ae2_params[0], ae1_params[3]) + ae2_params[1])\n",
        "    \n",
        "    classifier.fc3.weight = nn.Parameter(torch.matmul(ae3_params[0], ae2_params[2]))\n",
        "    classifier.fc3.bias = nn.Parameter(torch.matmul(ae3_params[0], ae2_params[3]) + ae3_params[1])\n",
        "    \n",
        "    classifier.fc4.weight = nn.Parameter(ae3_params[2])\n",
        "    classifier.fc4.bias = nn.Parameter(ae3_params[3])"
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "htcYHkZvb9EX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "criterion = nn.CrossEntropyLoss()\n",
        "optimizer = optim.SGD(classifier.parameters(), lr=0.001, momentum=0.9)\n",
        "classifier = classifier.to(device)"
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gGqhu8MUb9Ed",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "b5d6d98f-179e-40a5-cb7e-7cfb3751ecf5"
      },
      "source": [
        "old_loss = np.inf\n",
        "\n",
        "max_epoch = 500\n",
        "losses = []\n",
        "for epoch in range(max_epoch):\n",
        "\n",
        "    running_loss = 0.0\n",
        "    \n",
        "    for data in trainloader:\n",
        "        \n",
        "        X, y = data[0].to(device), data[1].to(device)\n",
        "        \n",
        "        optimizer.zero_grad()\n",
        "        \n",
        "        # Forward\n",
        "        y_hat = classifier(X)\n",
        "        \n",
        "        # Calculate Loss (Cross Entropy)\n",
        "        loss = criterion(y_hat, y)\n",
        "        \n",
        "        # Backpropagation\n",
        "        loss.backward()\n",
        "        \n",
        "        # Update Parameters\n",
        "        optimizer.step()\n",
        "        \n",
        "        running_loss += loss.item()*len(X)/train_size\n",
        "    \n",
        "    print('Epoch', epoch+1, ': Loss =', running_loss)\n",
        "    losses.append(running_loss)\n",
        "    if abs(running_loss-old_loss)/running_loss < 0.1 and running_loss<0.075:\n",
        "        print('Converged')\n",
        "        break\n",
        "    \n",
        "    old_loss = running_loss\n",
        "\n",
        "print('Finished Training')\n",
        "plt.plot(losses)\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Iter Number')\n",
        "plt.title('Convergence monitor plot')\n",
        "plt.show()"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1 : Loss = 1.6091626394878733\n",
            "Epoch 2 : Loss = 1.6014823019504556\n",
            "Epoch 3 : Loss = 1.6001403602686795\n",
            "Epoch 4 : Loss = 1.6000839146700767\n",
            "Epoch 5 : Loss = 1.5998983437364753\n",
            "Epoch 6 : Loss = 1.6006035262888128\n",
            "Epoch 7 : Loss = 1.600783589211377\n",
            "Epoch 8 : Loss = 1.599953304637562\n",
            "Epoch 9 : Loss = 1.600238667293028\n",
            "Epoch 10 : Loss = 1.6000705632296475\n",
            "Epoch 11 : Loss = 1.6002734059637238\n",
            "Epoch 12 : Loss = 1.6001130207018421\n",
            "Epoch 13 : Loss = 1.6001790301366283\n",
            "Epoch 14 : Loss = 1.600510041822087\n",
            "Epoch 15 : Loss = 1.6001043428074229\n",
            "Epoch 16 : Loss = 1.600325454365123\n",
            "Epoch 17 : Loss = 1.5999810587276113\n",
            "Epoch 18 : Loss = 1.6009076291864568\n",
            "Epoch 19 : Loss = 1.6001184555617245\n",
            "Epoch 20 : Loss = 1.6000924110412595\n",
            "Epoch 21 : Loss = 1.6006361625411294\n",
            "Epoch 22 : Loss = 1.6001604849641975\n",
            "Epoch 23 : Loss = 1.5997536209496586\n",
            "Epoch 24 : Loss = 1.600358228791844\n",
            "Epoch 25 : Loss = 1.6003575026988985\n",
            "Epoch 26 : Loss = 1.6000937629829755\n",
            "Epoch 27 : Loss = 1.6002937853336334\n",
            "Epoch 28 : Loss = 1.6008938632228158\n",
            "Epoch 29 : Loss = 1.6004648208618164\n",
            "Epoch 30 : Loss = 1.600474178791046\n",
            "Epoch 31 : Loss = 1.5998679318211297\n",
            "Epoch 32 : Loss = 1.6003694940697064\n",
            "Epoch 33 : Loss = 1.6000956459478899\n",
            "Epoch 34 : Loss = 1.5999844345179473\n",
            "Epoch 35 : Loss = 1.5999247513034127\n",
            "Epoch 36 : Loss = 1.5999049435962331\n",
            "Epoch 37 : Loss = 1.6004405428062785\n",
            "Epoch 38 : Loss = 1.6001566865227432\n",
            "Epoch 39 : Loss = 1.6001739691604266\n",
            "Epoch 40 : Loss = 1.5997282537547022\n",
            "Epoch 41 : Loss = 1.599966978484934\n",
            "Epoch 42 : Loss = 1.5997141951864418\n",
            "Epoch 43 : Loss = 1.5997802425514567\n",
            "Epoch 44 : Loss = 1.6000670194625854\n",
            "Epoch 45 : Loss = 1.6000554534521967\n",
            "Epoch 46 : Loss = 1.6000265804204075\n",
            "Epoch 47 : Loss = 1.5997895110737188\n",
            "Epoch 48 : Loss = 1.6001034514470536\n",
            "Epoch 49 : Loss = 1.6003532111644745\n",
            "Epoch 50 : Loss = 1.5999428006735716\n",
            "Epoch 51 : Loss = 1.6004407270388172\n",
            "Epoch 52 : Loss = 1.6006638949567624\n",
            "Epoch 53 : Loss = 1.6003001278096975\n",
            "Epoch 54 : Loss = 1.599965214729309\n",
            "Epoch 55 : Loss = 1.6004621711644262\n",
            "Epoch 56 : Loss = 1.5997576361352748\n",
            "Epoch 57 : Loss = 1.5998858782378118\n",
            "Epoch 58 : Loss = 1.6000664830207825\n",
            "Epoch 59 : Loss = 1.600045930255543\n",
            "Epoch 60 : Loss = 1.6003754355690696\n",
            "Epoch 61 : Loss = 1.6005424802953543\n",
            "Epoch 62 : Loss = 1.5998626242984424\n",
            "Epoch 63 : Loss = 1.5997010496529664\n",
            "Epoch 64 : Loss = 1.600390082055872\n",
            "Epoch 65 : Loss = 1.6001439473845742\n",
            "Epoch 66 : Loss = 1.599734837358648\n",
            "Epoch 67 : Loss = 1.6002401682463565\n",
            "Epoch 68 : Loss = 1.5999681570313198\n",
            "Epoch 69 : Loss = 1.5999070731076324\n",
            "Epoch 70 : Loss = 1.5998250327327035\n",
            "Epoch 71 : Loss = 1.5999296307563777\n",
            "Epoch 72 : Loss = 1.5998823886567897\n",
            "Epoch 73 : Loss = 1.5995690687136215\n",
            "Epoch 74 : Loss = 1.6001386073502628\n",
            "Epoch 75 : Loss = 1.6004884405569595\n",
            "Epoch 76 : Loss = 1.5995751348408787\n",
            "Epoch 77 : Loss = 1.5996626236222007\n",
            "Epoch 78 : Loss = 1.600032703443007\n",
            "Epoch 79 : Loss = 1.5999295250935988\n",
            "Epoch 80 : Loss = 1.60051197626374\n",
            "Epoch 81 : Loss = 1.5998855585401706\n",
            "Epoch 82 : Loss = 1.5999157835136757\n",
            "Epoch 83 : Loss = 1.5997457097877157\n",
            "Epoch 84 : Loss = 1.599645251577551\n",
            "Epoch 85 : Loss = 1.599606711756099\n",
            "Epoch 86 : Loss = 1.6000375828959725\n",
            "Epoch 87 : Loss = 1.5996581261808225\n",
            "Epoch 88 : Loss = 1.5998672815886414\n",
            "Epoch 89 : Loss = 1.5996005318381563\n",
            "Epoch 90 : Loss = 1.5994514118541372\n",
            "Epoch 91 : Loss = 1.5998667966235762\n",
            "Epoch 92 : Loss = 1.5999780920418831\n",
            "Epoch 93 : Loss = 1.6003325825387786\n",
            "Epoch 94 : Loss = 1.599865192716772\n",
            "Epoch 95 : Loss = 1.5999083952470259\n",
            "Epoch 96 : Loss = 1.5995137258009477\n",
            "Epoch 97 : Loss = 1.5998495383696123\n",
            "Epoch 98 : Loss = 1.599960216067054\n",
            "Epoch 99 : Loss = 1.6002009510993958\n",
            "Epoch 100 : Loss = 1.5994159741835163\n",
            "Epoch 101 : Loss = 1.5998660353097047\n",
            "Epoch 102 : Loss = 1.6009829965504732\n",
            "Epoch 103 : Loss = 1.5997356230562387\n",
            "Epoch 104 : Loss = 1.6000509668480265\n",
            "Epoch 105 : Loss = 1.5995433222163813\n",
            "Epoch 106 : Loss = 1.599883764982224\n",
            "Epoch 107 : Loss = 1.5998125834898518\n",
            "Epoch 108 : Loss = 1.5995939617807218\n",
            "Epoch 109 : Loss = 1.599614896557548\n",
            "Epoch 110 : Loss = 1.59946211901578\n",
            "Epoch 111 : Loss = 1.5996446663683108\n",
            "Epoch 112 : Loss = 1.5996954684907743\n",
            "Epoch 113 : Loss = 1.5995703258297664\n",
            "Epoch 114 : Loss = 1.5999151251532817\n",
            "Epoch 115 : Loss = 1.6001357761296355\n",
            "Epoch 116 : Loss = 1.5997538512403313\n",
            "Epoch 117 : Loss = 1.5993310809135437\n",
            "Epoch 118 : Loss = 1.6000824651934875\n",
            "Epoch 119 : Loss = 1.60046722130342\n",
            "Epoch 120 : Loss = 1.5997903021899138\n",
            "Epoch 121 : Loss = 1.599514056335796\n",
            "Epoch 122 : Loss = 1.5997753468426794\n",
            "Epoch 123 : Loss = 1.5994744923981752\n",
            "Epoch 124 : Loss = 1.6002862317995592\n",
            "Epoch 125 : Loss = 1.599422479217703\n",
            "Epoch 126 : Loss = 1.5998155610127882\n",
            "Epoch 127 : Loss = 1.5996474650773136\n",
            "Epoch 128 : Loss = 1.5995365652171047\n",
            "Epoch 129 : Loss = 1.5996295538815586\n",
            "Epoch 130 : Loss = 1.5994197780435744\n",
            "Epoch 131 : Loss = 1.5991846512664447\n",
            "Epoch 132 : Loss = 1.5994289138100362\n",
            "Epoch 133 : Loss = 1.5992067998105834\n",
            "Epoch 134 : Loss = 1.5994402739134703\n",
            "Epoch 135 : Loss = 1.5989471917802633\n",
            "Epoch 136 : Loss = 1.5992280380292374\n",
            "Epoch 137 : Loss = 1.5998860922726716\n",
            "Epoch 138 : Loss = 1.600285521962426\n",
            "Epoch 139 : Loss = 1.5994725633751263\n",
            "Epoch 140 : Loss = 1.5994397699832914\n",
            "Epoch 141 : Loss = 1.5989839976484126\n",
            "Epoch 142 : Loss = 1.5993882986632264\n",
            "Epoch 143 : Loss = 1.5994236821478065\n",
            "Epoch 144 : Loss = 1.599520718509501\n",
            "Epoch 145 : Loss = 1.599170511419123\n",
            "Epoch 146 : Loss = 1.5988354899666526\n",
            "Epoch 147 : Loss = 1.5990037565881554\n",
            "Epoch 148 : Loss = 1.5988542274995283\n",
            "Epoch 149 : Loss = 1.5991740199652587\n",
            "Epoch 150 : Loss = 1.598679369146174\n",
            "Epoch 151 : Loss = 1.5982876609672196\n",
            "Epoch 152 : Loss = 1.598579509691759\n",
            "Epoch 153 : Loss = 1.5986156680367216\n",
            "Epoch 154 : Loss = 1.599219785495238\n",
            "Epoch 155 : Loss = 1.5994168249043543\n",
            "Epoch 156 : Loss = 1.598489425399087\n",
            "Epoch 157 : Loss = 1.5985386209054422\n",
            "Epoch 158 : Loss = 1.5990174141797155\n",
            "Epoch 159 : Loss = 1.5988930328325792\n",
            "Epoch 160 : Loss = 1.5987456034530292\n",
            "Epoch 161 : Loss = 1.598146308552135\n",
            "Epoch 162 : Loss = 1.5980979908596382\n",
            "Epoch 163 : Loss = 1.5983619419011204\n",
            "Epoch 164 : Loss = 1.5981608629226687\n",
            "Epoch 165 : Loss = 1.5982531146569687\n",
            "Epoch 166 : Loss = 1.5976931750774384\n",
            "Epoch 167 : Loss = 1.5980326126922255\n",
            "Epoch 168 : Loss = 1.5980435338887302\n",
            "Epoch 169 : Loss = 1.5973860079591926\n",
            "Epoch 170 : Loss = 1.5973239513960749\n",
            "Epoch 171 : Loss = 1.597765191034837\n",
            "Epoch 172 : Loss = 1.5974533314054664\n",
            "Epoch 173 : Loss = 1.597431727431037\n",
            "Epoch 174 : Loss = 1.5966108631003986\n",
            "Epoch 175 : Loss = 1.5967332178896128\n",
            "Epoch 176 : Loss = 1.5966581891883505\n",
            "Epoch 177 : Loss = 1.5959763012149117\n",
            "Epoch 178 : Loss = 1.5965150432153183\n",
            "Epoch 179 : Loss = 1.5957787768407303\n",
            "Epoch 180 : Loss = 1.5959590023214167\n",
            "Epoch 181 : Loss = 1.596122026443482\n",
            "Epoch 182 : Loss = 1.5953194092620508\n",
            "Epoch 183 : Loss = 1.5945032780820672\n",
            "Epoch 184 : Loss = 1.5952278890392997\n",
            "Epoch 185 : Loss = 1.5952412106774072\n",
            "Epoch 186 : Loss = 1.5933110415935519\n",
            "Epoch 187 : Loss = 1.593556940555573\n",
            "Epoch 188 : Loss = 1.5931791202588512\n",
            "Epoch 189 : Loss = 1.5922262316400355\n",
            "Epoch 190 : Loss = 1.5915531678633252\n",
            "Epoch 191 : Loss = 1.5916181071238082\n",
            "Epoch 192 : Loss = 1.589831016280434\n",
            "Epoch 193 : Loss = 1.5889074883677745\n",
            "Epoch 194 : Loss = 1.5893346179615364\n",
            "Epoch 195 : Loss = 1.586725088683042\n",
            "Epoch 196 : Loss = 1.58557131615552\n",
            "Epoch 197 : Loss = 1.5850044109604575\n",
            "Epoch 198 : Loss = 1.5820050320842045\n",
            "Epoch 199 : Loss = 1.579311164942655\n",
            "Epoch 200 : Loss = 1.5790552903305404\n",
            "Epoch 201 : Loss = 1.5759695903821427\n",
            "Epoch 202 : Loss = 1.5706413903019643\n",
            "Epoch 203 : Loss = 1.5658870312300595\n",
            "Epoch 204 : Loss = 1.563180368054997\n",
            "Epoch 205 : Loss = 1.5545197318900714\n",
            "Epoch 206 : Loss = 1.5467037558555599\n",
            "Epoch 207 : Loss = 1.5355888876048\n",
            "Epoch 208 : Loss = 1.5242907784201885\n",
            "Epoch 209 : Loss = 1.510207062417811\n",
            "Epoch 210 : Loss = 1.4958039034496655\n",
            "Epoch 211 : Loss = 1.4788940007036386\n",
            "Epoch 212 : Loss = 1.4627347724004225\n",
            "Epoch 213 : Loss = 1.4485145319591866\n",
            "Epoch 214 : Loss = 1.4370498873970725\n",
            "Epoch 215 : Loss = 1.420290196483785\n",
            "Epoch 216 : Loss = 1.4129688956520776\n",
            "Epoch 217 : Loss = 1.3898667611859064\n",
            "Epoch 218 : Loss = 1.3848295022140849\n",
            "Epoch 219 : Loss = 1.3852869326418098\n",
            "Epoch 220 : Loss = 1.3629933406006207\n",
            "Epoch 221 : Loss = 1.3945917622609574\n",
            "Epoch 222 : Loss = 1.3496029160239476\n",
            "Epoch 223 : Loss = 1.3717215548862112\n",
            "Epoch 224 : Loss = 1.3409754417159336\n",
            "Epoch 225 : Loss = 1.3978255580772054\n",
            "Epoch 226 : Loss = 1.3839330375194552\n",
            "Epoch 227 : Loss = 1.3387523970820692\n",
            "Epoch 228 : Loss = 1.31192176992243\n",
            "Epoch 229 : Loss = 1.3147374499927866\n",
            "Epoch 230 : Loss = 1.3220836574381043\n",
            "Epoch 231 : Loss = 1.2917982258579945\n",
            "Epoch 232 : Loss = 1.2884614332155748\n",
            "Epoch 233 : Loss = 1.3012727742845362\n",
            "Epoch 234 : Loss = 1.2715272117744791\n",
            "Epoch 235 : Loss = 1.303255875002254\n",
            "Epoch 236 : Loss = 1.2859745567495173\n",
            "Epoch 237 : Loss = 1.2849227975715292\n",
            "Epoch 238 : Loss = 1.2593448622660204\n",
            "Epoch 239 : Loss = 1.2871172157200899\n",
            "Epoch 240 : Loss = 1.3205826634710482\n",
            "Epoch 241 : Loss = 1.2889717465097257\n",
            "Epoch 242 : Loss = 1.2596129937605427\n",
            "Epoch 243 : Loss = 1.2468933571468706\n",
            "Epoch 244 : Loss = 1.2677573751319542\n",
            "Epoch 245 : Loss = 1.250888529148969\n",
            "Epoch 246 : Loss = 1.2510305697267705\n",
            "Epoch 247 : Loss = 1.2268360609358007\n",
            "Epoch 248 : Loss = 1.3043247705156151\n",
            "Epoch 249 : Loss = 1.2751490257003084\n",
            "Epoch 250 : Loss = 1.2375105646523556\n",
            "Epoch 251 : Loss = 1.2264771624044937\n",
            "Epoch 252 : Loss = 1.2245421084490693\n",
            "Epoch 253 : Loss = 1.1976290995424443\n",
            "Epoch 254 : Loss = 1.185483122413809\n",
            "Epoch 255 : Loss = 1.210127678784457\n",
            "Epoch 256 : Loss = 1.2019673030484808\n",
            "Epoch 257 : Loss = 1.198393071239645\n",
            "Epoch 258 : Loss = 1.179591728882356\n",
            "Epoch 259 : Loss = 1.3214674659750676\n",
            "Epoch 260 : Loss = 1.2291681116277517\n",
            "Epoch 261 : Loss = 1.1745208285071633\n",
            "Epoch 262 : Loss = 1.1737581897865643\n",
            "Epoch 263 : Loss = 1.2533397132700144\n",
            "Epoch 264 : Loss = 1.1805776953697205\n",
            "Epoch 265 : Loss = 1.1504603786901997\n",
            "Epoch 266 : Loss = 1.2097943682562224\n",
            "Epoch 267 : Loss = 1.2466403137553819\n",
            "Epoch 268 : Loss = 1.1498007693073966\n",
            "Epoch 269 : Loss = 1.1592292325063183\n",
            "Epoch 270 : Loss = 1.1423453363505278\n",
            "Epoch 271 : Loss = 1.1342309483072974\n",
            "Epoch 272 : Loss = 1.1252637573263857\n",
            "Epoch 273 : Loss = 1.1843150921843268\n",
            "Epoch 274 : Loss = 1.1879978667606008\n",
            "Epoch 275 : Loss = 1.1232431436126882\n",
            "Epoch 276 : Loss = 1.1520675894888965\n",
            "Epoch 277 : Loss = 1.1078714592890302\n",
            "Epoch 278 : Loss = 1.304250014099208\n",
            "Epoch 279 : Loss = 1.1119324104352428\n",
            "Epoch 280 : Loss = 1.0924522212960501\n",
            "Epoch 281 : Loss = 1.0812652354890648\n",
            "Epoch 282 : Loss = 1.153057325970043\n",
            "Epoch 283 : Loss = 1.1132616062055936\n",
            "Epoch 284 : Loss = 1.1397359032522547\n",
            "Epoch 285 : Loss = 1.0570142932913518\n",
            "Epoch 286 : Loss = 1.0535839904438362\n",
            "Epoch 287 : Loss = 1.039613437923518\n",
            "Epoch 288 : Loss = 1.1037091667001897\n",
            "Epoch 289 : Loss = 1.0670398013158275\n",
            "Epoch 290 : Loss = 1.0696315101601859\n",
            "Epoch 291 : Loss = 1.0159102068705992\n",
            "Epoch 292 : Loss = 1.0112499174746599\n",
            "Epoch 293 : Loss = 0.981730031696233\n",
            "Epoch 294 : Loss = 1.013111384077506\n",
            "Epoch 295 : Loss = 1.0379750119014222\n",
            "Epoch 296 : Loss = 0.9552085223523054\n",
            "Epoch 297 : Loss = 0.9195519428361546\n",
            "Epoch 298 : Loss = 0.9154205918312075\n",
            "Epoch 299 : Loss = 0.9132885377515448\n",
            "Epoch 300 : Loss = 0.8655287027359009\n",
            "Epoch 301 : Loss = 0.8473641208627003\n",
            "Epoch 302 : Loss = 0.8907787718556148\n",
            "Epoch 303 : Loss = 0.8802883733402598\n",
            "Epoch 304 : Loss = 0.8074258091774855\n",
            "Epoch 305 : Loss = 0.8747777925296262\n",
            "Epoch 306 : Loss = 0.8389109725301915\n",
            "Epoch 307 : Loss = 0.833910950205543\n",
            "Epoch 308 : Loss = 0.8773131235079332\n",
            "Epoch 309 : Loss = 0.7814900888638063\n",
            "Epoch 310 : Loss = 0.7541807849298825\n",
            "Epoch 311 : Loss = 0.7349061478268015\n",
            "Epoch 312 : Loss = 0.7598509761420165\n",
            "Epoch 313 : Loss = 0.7081875834952701\n",
            "Epoch 314 : Loss = 0.7516365911472929\n",
            "Epoch 315 : Loss = 0.8159352750940759\n",
            "Epoch 316 : Loss = 0.6810852194374256\n",
            "Epoch 317 : Loss = 0.7018856033682825\n",
            "Epoch 318 : Loss = 0.702873965555971\n",
            "Epoch 319 : Loss = 0.7298091433264995\n",
            "Epoch 320 : Loss = 0.6931838826699692\n",
            "Epoch 321 : Loss = 0.6845364618030463\n",
            "Epoch 322 : Loss = 0.789758102460341\n",
            "Epoch 323 : Loss = 0.6685427982698786\n",
            "Epoch 324 : Loss = 0.656752783466469\n",
            "Epoch 325 : Loss = 0.840320224789056\n",
            "Epoch 326 : Loss = 0.8002787970683789\n",
            "Epoch 327 : Loss = 0.6075768971985036\n",
            "Epoch 328 : Loss = 0.6436256685040217\n",
            "Epoch 329 : Loss = 0.6046471318060703\n",
            "Epoch 330 : Loss = 0.6431039111180739\n",
            "Epoch 331 : Loss = 0.5656865259463137\n",
            "Epoch 332 : Loss = 0.6429996910420333\n",
            "Epoch 333 : Loss = 0.5485696636817672\n",
            "Epoch 334 : Loss = 0.6013370420445096\n",
            "Epoch 335 : Loss = 0.5207481309771539\n",
            "Epoch 336 : Loss = 0.5016619420864366\n",
            "Epoch 337 : Loss = 0.6185113408348776\n",
            "Epoch 338 : Loss = 0.6236194907264275\n",
            "Epoch 339 : Loss = 0.5347477942705154\n",
            "Epoch 340 : Loss = 0.5126422426917335\n",
            "Epoch 341 : Loss = 0.5875313600355931\n",
            "Epoch 342 : Loss = 0.483127430758693\n",
            "Epoch 343 : Loss = 0.46001377972689533\n",
            "Epoch 344 : Loss = 0.4656405760483308\n",
            "Epoch 345 : Loss = 0.44477182762189343\n",
            "Epoch 346 : Loss = 0.5056154470552098\n",
            "Epoch 347 : Loss = 0.48717696761543106\n",
            "Epoch 348 : Loss = 0.4568426832556723\n",
            "Epoch 349 : Loss = 0.4775599366561934\n",
            "Epoch 350 : Loss = 0.5008803893896666\n",
            "Epoch 351 : Loss = 0.4920226799493486\n",
            "Epoch 352 : Loss = 0.4141888079995459\n",
            "Epoch 353 : Loss = 0.6165962310677225\n",
            "Epoch 354 : Loss = 0.427717016840523\n",
            "Epoch 355 : Loss = 0.4264117228713901\n",
            "Epoch 356 : Loss = 0.3821519755504349\n",
            "Epoch 357 : Loss = 0.35030021078207274\n",
            "Epoch 358 : Loss = 0.3339082852683284\n",
            "Epoch 359 : Loss = 0.5179580375552177\n",
            "Epoch 360 : Loss = 0.42199834300713107\n",
            "Epoch 361 : Loss = 0.4864254197613761\n",
            "Epoch 362 : Loss = 0.4496505270627412\n",
            "Epoch 363 : Loss = 0.42441875385967176\n",
            "Epoch 364 : Loss = 0.3429094583473422\n",
            "Epoch 365 : Loss = 0.4052463664927266\n",
            "Epoch 366 : Loss = 0.30499373139305547\n",
            "Epoch 367 : Loss = 0.3652354539795356\n",
            "Epoch 368 : Loss = 0.35803547196767554\n",
            "Epoch 369 : Loss = 0.30119205774231406\n",
            "Epoch 370 : Loss = 0.35696076923473324\n",
            "Epoch 371 : Loss = 0.29346585206010123\n",
            "Epoch 372 : Loss = 0.3315615579485893\n",
            "Epoch 373 : Loss = 0.32280106707052747\n",
            "Epoch 374 : Loss = 0.31292638335038314\n",
            "Epoch 375 : Loss = 0.2713948237625035\n",
            "Epoch 376 : Loss = 0.369640994139693\n",
            "Epoch 377 : Loss = 0.43814089369367465\n",
            "Epoch 378 : Loss = 0.31727060066028073\n",
            "Epoch 379 : Loss = 0.33974123542959045\n",
            "Epoch 380 : Loss = 0.3017428747632287\n",
            "Epoch 381 : Loss = 0.5492817370051688\n",
            "Epoch 382 : Loss = 0.4746801572089846\n",
            "Epoch 383 : Loss = 0.35320886901833787\n",
            "Epoch 384 : Loss = 0.2457951844077218\n",
            "Epoch 385 : Loss = 0.2391557903452353\n",
            "Epoch 386 : Loss = 0.2269501379606399\n",
            "Epoch 387 : Loss = 0.24182025885040107\n",
            "Epoch 388 : Loss = 0.22672198662026363\n",
            "Epoch 389 : Loss = 0.2142499544742432\n",
            "Epoch 390 : Loss = 0.19334570792588326\n",
            "Epoch 391 : Loss = 0.20654678920453243\n",
            "Epoch 392 : Loss = 0.20148227296092291\n",
            "Epoch 393 : Loss = 0.20234387351030647\n",
            "Epoch 394 : Loss = 0.19054875759915874\n",
            "Epoch 395 : Loss = 0.3194865242324093\n",
            "Epoch 396 : Loss = 0.3956227912144227\n",
            "Epoch 397 : Loss = 0.5343156795610082\n",
            "Epoch 398 : Loss = 0.28367651998996735\n",
            "Epoch 399 : Loss = 0.1866954692046751\n",
            "Epoch 400 : Loss = 0.17740025354379957\n",
            "Epoch 401 : Loss = 0.20137865299528293\n",
            "Epoch 402 : Loss = 0.1871748099272901\n",
            "Epoch 403 : Loss = 0.39216266572475444\n",
            "Epoch 404 : Loss = 0.27606002817099745\n",
            "Epoch 405 : Loss = 0.5092959900132634\n",
            "Epoch 406 : Loss = 0.1740875779227777\n",
            "Epoch 407 : Loss = 0.16673066678710957\n",
            "Epoch 408 : Loss = 0.19122993709011507\n",
            "Epoch 409 : Loss = 0.14831917631355201\n",
            "Epoch 410 : Loss = 0.2532585554502227\n",
            "Epoch 411 : Loss = 0.1664108615368604\n",
            "Epoch 412 : Loss = 0.17752300160513687\n",
            "Epoch 413 : Loss = 0.15502937290478838\n",
            "Epoch 414 : Loss = 0.14462611705742098\n",
            "Epoch 415 : Loss = 0.14039686068215157\n",
            "Epoch 416 : Loss = 0.1304159784181551\n",
            "Epoch 417 : Loss = 0.1685914420946078\n",
            "Epoch 418 : Loss = 0.1405213795263659\n",
            "Epoch 419 : Loss = 0.13327418665655633\n",
            "Epoch 420 : Loss = 0.154878346842121\n",
            "Epoch 421 : Loss = 0.2812610426917672\n",
            "Epoch 422 : Loss = 0.2085951741107486\n",
            "Epoch 423 : Loss = 0.16848170884292238\n",
            "Epoch 424 : Loss = 0.13638939039612355\n",
            "Epoch 425 : Loss = 0.16669196982614023\n",
            "Epoch 426 : Loss = 0.11028547009283848\n",
            "Epoch 427 : Loss = 0.09892414247786455\n",
            "Epoch 428 : Loss = 0.14524760622192512\n",
            "Epoch 429 : Loss = 0.15001844191415745\n",
            "Epoch 430 : Loss = 0.09443310563537209\n",
            "Epoch 431 : Loss = 0.09957079432735388\n",
            "Epoch 432 : Loss = 0.0876425461504947\n",
            "Epoch 433 : Loss = 0.0837758300741288\n",
            "Epoch 434 : Loss = 0.0991799640469253\n",
            "Epoch 435 : Loss = 0.09806992816315457\n",
            "Epoch 436 : Loss = 0.11500371509993616\n",
            "Epoch 437 : Loss = 0.16556050289760937\n",
            "Epoch 438 : Loss = 0.17932128813117745\n",
            "Epoch 439 : Loss = 0.18192661549388014\n",
            "Epoch 440 : Loss = 0.1259010688147762\n",
            "Epoch 441 : Loss = 0.10781242359768259\n",
            "Epoch 442 : Loss = 0.13261446797034954\n",
            "Epoch 443 : Loss = 0.10469184096225283\n",
            "Epoch 444 : Loss = 0.06582693179899997\n",
            "Epoch 445 : Loss = 0.06336983217095786\n",
            "Converged\n",
            "Finished Training\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d-5XLFwwMwXd",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 153
        },
        "outputId": "d2f6ef85-a3e8-4e2b-e0c6-c8621c6f78b3"
      },
      "source": [
        "with torch.no_grad():\n",
        "    \n",
        "    train_loss = 0.0\n",
        "    y_train = []\n",
        "    y_train_pred = []\n",
        "\n",
        "    for data in trainloader:\n",
        "\n",
        "        X, y = data[0].to(device), data[1].to(device)\n",
        "        y_hat = classifier(X)      \n",
        "        train_loss += criterion(y_hat, y)*len(X)/train_size\n",
        "        \n",
        "        y_train.extend(list(y.detach().cpu().numpy()))\n",
        "        y_train_pred.extend(list(torch.argmax(y_hat, axis=1).detach().cpu().numpy()))\n",
        "\n",
        "print('Train Loss =', train_loss.item())\n",
        "print(pd.DataFrame(confusion_matrix(y_train, y_train_pred)))\n",
        "\n",
        "acc = accuracy_score(y_train, y_train_pred)\n",
        "prec = precision_score(y_train, y_train_pred, average='macro')\n",
        "f1 = f1_score(y_train, y_train_pred, average='macro')\n",
        "\n",
        "print('Train Accuracy =', acc, 'Train Precision =', prec, 'Train F1 =', f1)"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train Loss = 0.05523950979113579\n",
            "     0    1    2    3    4\n",
            "0  282    0    0    0    0\n",
            "1    0  294    1    0    1\n",
            "2    1    0  210    0    0\n",
            "3    0    0    0  280    7\n",
            "4    0    0    0    4  328\n",
            "Train Accuracy = 0.9900568181818182 Train Precision = 0.9907666127501475 Train F1 = 0.9905744802603339\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yVYcIfixb9Eg",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 221
        },
        "outputId": "c15717cd-8b02-4ef4-e5e3-18dbdfb5d768"
      },
      "source": [
        "with torch.no_grad():\n",
        "    \n",
        "    test_loss = 0.0\n",
        "    y_test = []\n",
        "    y_test_pred = []\n",
        "\n",
        "    for data in testloader:\n",
        "\n",
        "        X, y = data[0].to(device), data[1].to(device)\n",
        "        y_hat = classifier(X)      \n",
        "        test_loss += criterion(y_hat, y)*len(X)/test_size\n",
        "        \n",
        "        y_test.extend(list(y.detach().cpu().numpy()))\n",
        "        y_test_pred.extend(list(torch.argmax(y_hat, axis=1).detach().cpu().numpy()))\n",
        "\n",
        "print('Test Loss =', test_loss.item())\n",
        "pd.DataFrame(confusion_matrix(y_test, y_test_pred))"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Test Loss = 1.577786922454834\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "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>0</th>\n",
              "      <th>1</th>\n",
              "      <th>2</th>\n",
              "      <th>3</th>\n",
              "      <th>4</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>45</td>\n",
              "      <td>16</td>\n",
              "      <td>8</td>\n",
              "      <td>3</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>11</td>\n",
              "      <td>45</td>\n",
              "      <td>5</td>\n",
              "      <td>6</td>\n",
              "      <td>11</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "      <td>28</td>\n",
              "      <td>14</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "      <td>14</td>\n",
              "      <td>41</td>\n",
              "      <td>14</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1</td>\n",
              "      <td>7</td>\n",
              "      <td>2</td>\n",
              "      <td>9</td>\n",
              "      <td>59</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "    0   1   2   3   4\n",
              "0  45  16   8   3   2\n",
              "1  11  45   5   6  11\n",
              "2   1   5  28  14   1\n",
              "3   1   3  14  41  14\n",
              "4   1   7   2   9  59"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 30
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jr5SF1L7Tmvh",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "04c20e24-be3a-4100-f98d-f697b02f3d32"
      },
      "source": [
        "acc = accuracy_score(y_test, y_test_pred)\n",
        "prec = precision_score(y_test, y_test_pred, average='macro')\n",
        "f1 = f1_score(y_test, y_test_pred, average='macro')\n",
        "\n",
        "print('Test Accuracy =', acc, 'Test Precision =', prec, 'Test F1 =', f1)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Test Accuracy = 0.6193181818181818 Test Precision = 0.6171699905793562 Test F1 = 0.6132409102598597\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5Qk2zBLeQXsY",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "torch.save(classifier, '/content/drive/My Drive/A2Q2_model.pt')"
      ],
      "execution_count": 32,
      "outputs": []
    }
  ]
}