Classification using PyTorch/ML_HW4_Answer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "MT-ucU8DVPt8"
   },
   "source": [
    "# Bank Marketing Classification\n",
    "\n",
    "In this part of the assignment we want to do a classification task using PyTorch on bank marketing dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "rZlNNmgqYkV4"
   },
   "source": [
    "#Import Library"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "id": "NOh9eJ4pYnu2"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import pandas as pd\n",
    "import tqdm\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler, LabelEncoder\n",
    "from torch.utils.data import DataLoader, TensorDataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "EJaMVgX1XcZm"
   },
   "source": [
    "#Load Dataset\n",
    "Here, you should download and load the dataset. After that, you should split train dataset to train and validation sets. Finally, define the dataloaders for `train`, `validation` and `test`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "9YSstSPUXwu2"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>age</th>\n",
       "      <th>job</th>\n",
       "      <th>marital</th>\n",
       "      <th>education</th>\n",
       "      <th>default</th>\n",
       "      <th>balance</th>\n",
       "      <th>housing</th>\n",
       "      <th>loan</th>\n",
       "      <th>contact</th>\n",
       "      <th>day</th>\n",
       "      <th>month</th>\n",
       "      <th>duration</th>\n",
       "      <th>campaign</th>\n",
       "      <th>pdays</th>\n",
       "      <th>previous</th>\n",
       "      <th>poutcome</th>\n",
       "      <th>y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>58</td>\n",
       "      <td>management</td>\n",
       "      <td>married</td>\n",
       "      <td>tertiary</td>\n",
       "      <td>no</td>\n",
       "      <td>2143</td>\n",
       "      <td>yes</td>\n",
       "      <td>no</td>\n",
       "      <td>unknown</td>\n",
       "      <td>5</td>\n",
       "      <td>may</td>\n",
       "      <td>261</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>44</td>\n",
       "      <td>technician</td>\n",
       "      <td>single</td>\n",
       "      <td>secondary</td>\n",
       "      <td>no</td>\n",
       "      <td>29</td>\n",
       "      <td>yes</td>\n",
       "      <td>no</td>\n",
       "      <td>unknown</td>\n",
       "      <td>5</td>\n",
       "      <td>may</td>\n",
       "      <td>151</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>33</td>\n",
       "      <td>entrepreneur</td>\n",
       "      <td>married</td>\n",
       "      <td>secondary</td>\n",
       "      <td>no</td>\n",
       "      <td>2</td>\n",
       "      <td>yes</td>\n",
       "      <td>yes</td>\n",
       "      <td>unknown</td>\n",
       "      <td>5</td>\n",
       "      <td>may</td>\n",
       "      <td>76</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>47</td>\n",
       "      <td>blue-collar</td>\n",
       "      <td>married</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "      <td>1506</td>\n",
       "      <td>yes</td>\n",
       "      <td>no</td>\n",
       "      <td>unknown</td>\n",
       "      <td>5</td>\n",
       "      <td>may</td>\n",
       "      <td>92</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>33</td>\n",
       "      <td>unknown</td>\n",
       "      <td>single</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "      <td>1</td>\n",
       "      <td>no</td>\n",
       "      <td>no</td>\n",
       "      <td>unknown</td>\n",
       "      <td>5</td>\n",
       "      <td>may</td>\n",
       "      <td>198</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>unknown</td>\n",
       "      <td>no</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   age           job  marital  education default  balance housing loan  \\\n",
       "0   58    management  married   tertiary      no     2143     yes   no   \n",
       "1   44    technician   single  secondary      no       29     yes   no   \n",
       "2   33  entrepreneur  married  secondary      no        2     yes  yes   \n",
       "3   47   blue-collar  married    unknown      no     1506     yes   no   \n",
       "4   33       unknown   single    unknown      no        1      no   no   \n",
       "\n",
       "   contact  day month  duration  campaign  pdays  previous poutcome   y  \n",
       "0  unknown    5   may       261         1     -1         0  unknown  no  \n",
       "1  unknown    5   may       151         1     -1         0  unknown  no  \n",
       "2  unknown    5   may        76         1     -1         0  unknown  no  \n",
       "3  unknown    5   may        92         1     -1         0  unknown  no  \n",
       "4  unknown    5   may       198         1     -1         0  unknown  no  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv(\"./bank-full.csv\", sep=\";\")\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Keip2OUFX8qk"
   },
   "source": [
    "#Normalize data\n",
    "normalize numerical data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "id": "e-Wk_AoiYfo4"
   },
   "outputs": [],
   "source": [
    "scaler = StandardScaler()\n",
    "encoder = LabelEncoder()\n",
    "\n",
    "conts_cols = ['age', 'balance', 'day', 'duration', 'campaign', 'pdays', 'previous']\n",
    "df[conts_cols] = scaler.fit_transform(df[conts_cols])\n",
    "\n",
    "categories = ['job', 'marital', 'education', 'default', 'housing', 'loan', 'contact', 'month', 'poutcome']\n",
    "df = pd.get_dummies(df, columns=categories)\n",
    "\n",
    "df['y'] = encoder.fit_transform(df['y'])\n",
    "\n",
    "X = df.drop('y', axis=1)\n",
    "y = df['y']\n",
    "X = X.astype(np.float32)\n",
    "y = y.astype(np.float32)\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\n",
    "X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.25, random_state=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "8TnDg_OtYrWl"
   },
   "source": [
    "# Create Data Loader\n",
    "create data loader for pytorch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "id": "yIdgxZMUYqzE"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of batches in train loader:  335\n",
      "Number of batches in val loader:  112\n",
      "Number of batches in test loader:  112\n"
     ]
    }
   ],
   "source": [
    "X_train = torch.tensor(X_train.values, dtype=torch.float32)\n",
    "y_train = torch.tensor(y_train.values, dtype=torch.float32)\n",
    "X_val = torch.tensor(X_val.values, dtype=torch.float32)\n",
    "y_val = torch.tensor(y_val.values, dtype=torch.float32)\n",
    "X_test = torch.tensor(X_test.values, dtype=torch.float32)\n",
    "y_test = torch.tensor(y_test.values, dtype=torch.float32)\n",
    "\n",
    "y_train = y_train.reshape(-1, 1)\n",
    "y_val = y_val.reshape(-1, 1)\n",
    "y_test = y_test.reshape(-1, 1)\n",
    "\n",
    "train_dataset = TensorDataset(X_train, y_train)\n",
    "val_dataset = TensorDataset(X_val, y_val)\n",
    "test_dataset = TensorDataset(X_test, y_test)\n",
    "\n",
    "train_loader = DataLoader(train_dataset, batch_size=81, shuffle=True)\n",
    "val_loader = DataLoader(val_dataset, batch_size=81, shuffle=False)\n",
    "test_loader = DataLoader(test_dataset, batch_size=81, shuffle=False)\n",
    "\n",
    "print(\"Number of batches in train loader: \", len(train_loader))\n",
    "print(\"Number of batches in val loader: \", len(val_loader))\n",
    "print(\"Number of batches in test loader: \", len(test_loader))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "GBtQo2PBY9zS"
   },
   "source": [
    "#Model\n",
    "define your model in here"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "id": "C5k1GGexZCGD"
   },
   "outputs": [],
   "source": [
    "class BankMarketingNet(nn.Module):\n",
    "    def __init__(self, in_dim):\n",
    "        super().__init__()\n",
    "        self.relu = nn.ReLU()\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        self.fc1 = nn.Linear(in_dim, 128)\n",
    "        self.fc2 = nn.Linear(128, 32)\n",
    "        self.fc3 = nn.Linear(32, 4)\n",
    "        self.fc4 = nn.Linear(4, 1)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.relu(self.fc1(x))\n",
    "        x = self.relu(self.fc2(x))\n",
    "        x = self.relu(self.fc3(x))\n",
    "        x = self.sigmoid(self.fc4(x))\n",
    "        return x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "P_ef2VVkZHpj"
   },
   "source": [
    "#Train your model\n",
    "Define `criterion` and `optimizer`.\n",
    "Train your model\n",
    "\n",
    "Tasks:\n",
    "- [ ] Things that are needed to be printed in each epoch:\n",
    "  - Number of epoch\n",
    "  - Train loss\n",
    "  - Validation loss\n",
    "- [ ] save train/validation loss (of each epoch) in an array for later usage"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "7BC2acuYZH0j",
    "outputId": "813dc792-19f9-47ea-f65d-ceb1553aef10",
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 1/25,\n",
      "Training Loss:\t 0.29162491505252419177,\n",
      "Validation Loss: 0.22682770429390428091\n",
      "#####################################################################################################\n",
      "Epoch: 2/25,\n",
      "Training Loss:\t 0.22150089788759236287,\n",
      "Validation Loss: 0.22074102361172970310\n",
      "#####################################################################################################\n",
      "Epoch: 3/25,\n",
      "Training Loss:\t 0.21570953433837569690,\n",
      "Validation Loss: 0.21379126458074687589\n",
      "#####################################################################################################\n",
      "Epoch: 4/25,\n",
      "Training Loss:\t 0.21091715633799956731,\n",
      "Validation Loss: 0.21231421534844638344\n",
      "#####################################################################################################\n",
      "Epoch: 5/25,\n",
      "Training Loss:\t 0.20654355185804734729,\n",
      "Validation Loss: 0.21032048167330821831\n",
      "#####################################################################################################\n",
      "Epoch: 6/25,\n",
      "Training Loss:\t 0.20355562234140422984,\n",
      "Validation Loss: 0.20735398608283958644\n",
      "#####################################################################################################\n",
      "Epoch: 7/25,\n",
      "Training Loss:\t 0.19941466152371359333,\n",
      "Validation Loss: 0.20575050070051253970\n",
      "#####################################################################################################\n",
      "Epoch: 8/25,\n",
      "Training Loss:\t 0.19610508368922213718,\n",
      "Validation Loss: 0.20885506620866697847\n",
      "#####################################################################################################\n",
      "Epoch: 9/25,\n",
      "Training Loss:\t 0.19508753292752009645,\n",
      "Validation Loss: 0.20500052703004367771\n",
      "#####################################################################################################\n",
      "Epoch: 10/25,\n",
      "Training Loss:\t 0.19160779326160418123,\n",
      "Validation Loss: 0.20518924086277423591\n",
      "#####################################################################################################\n",
      "Epoch: 11/25,\n",
      "Training Loss:\t 0.18971726743784975144,\n",
      "Validation Loss: 0.20512062179662568506\n",
      "#####################################################################################################\n",
      "Epoch: 12/25,\n",
      "Training Loss:\t 0.18739475922160933608,\n",
      "Validation Loss: 0.20505033579532008781\n",
      "#####################################################################################################\n",
      "Epoch: 13/25,\n",
      "Training Loss:\t 0.18648784366258938650,\n",
      "Validation Loss: 0.20644089090128014963\n",
      "#####################################################################################################\n",
      "Epoch: 14/25,\n",
      "Training Loss:\t 0.18380775187453768371,\n",
      "Validation Loss: 0.20594358979152699307\n",
      "#####################################################################################################\n",
      "Epoch: 15/25,\n",
      "Training Loss:\t 0.18286593000675838150,\n",
      "Validation Loss: 0.20449370033074942832\n",
      "#####################################################################################################\n",
      "Epoch: 16/25,\n",
      "Training Loss:\t 0.18153180199654866955,\n",
      "Validation Loss: 0.20975007028544509491\n",
      "#####################################################################################################\n",
      "Epoch: 17/25,\n",
      "Training Loss:\t 0.17989298789289193126,\n",
      "Validation Loss: 0.20605335437013522215\n",
      "#####################################################################################################\n",
      "Epoch: 18/25,\n",
      "Training Loss:\t 0.17823953479380108544,\n",
      "Validation Loss: 0.20799838483343474627\n",
      "#####################################################################################################\n",
      "Epoch: 19/25,\n",
      "Training Loss:\t 0.17692467458570881877,\n",
      "Validation Loss: 0.20821710724158640216\n",
      "#####################################################################################################\n",
      "Epoch: 20/25,\n",
      "Training Loss:\t 0.17528069379733901556,\n",
      "Validation Loss: 0.20873243026509535891\n",
      "#####################################################################################################\n",
      "Epoch: 21/25,\n",
      "Training Loss:\t 0.17429568610872822609,\n",
      "Validation Loss: 0.20916711715530225679\n",
      "#####################################################################################################\n",
      "Epoch: 22/25,\n",
      "Training Loss:\t 0.17208806315576627433,\n",
      "Validation Loss: 0.21500788695798778161\n",
      "#####################################################################################################\n",
      "Epoch: 23/25,\n",
      "Training Loss:\t 0.17147489143974273262,\n",
      "Validation Loss: 0.21479139533870331147\n",
      "#####################################################################################################\n",
      "Epoch: 24/25,\n",
      "Training Loss:\t 0.16986413559596430511,\n",
      "Validation Loss: 0.21316677198436928542\n",
      "#####################################################################################################\n",
      "Epoch: 25/25,\n",
      "Training Loss:\t 0.16794302896426541172,\n",
      "Validation Loss: 0.21418436622591943697\n",
      "#####################################################################################################\n"
     ]
    }
   ],
   "source": [
    "model = BankMarketingNet(X_train.shape[1])\n",
    "\n",
    "optimizer = optim.Adam(model.parameters(), lr=0.0005)\n",
    "criterion = nn.BCELoss()\n",
    "\n",
    "train_losses = []\n",
    "val_losses = []\n",
    "for epoch in range(25):\n",
    "    model.train()\n",
    "    train_loss = 0\n",
    "    for X_batch, y_batch in train_loader:\n",
    "        optimizer.zero_grad()\n",
    "        y_pred = model(X_batch)\n",
    "        loss = criterion(y_pred, y_batch)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        train_loss += loss.item() * len(X_batch)\n",
    "    train_loss /= len(train_loader.dataset)\n",
    "    train_losses.append(train_loss)\n",
    "\n",
    "    model.eval()\n",
    "    val_loss = 0\n",
    "    with torch.no_grad():\n",
    "        for X_batch, y_batch in val_loader:\n",
    "            y_pred = model(X_batch)\n",
    "            loss = criterion(y_pred, y_batch)\n",
    "            val_loss += loss.item() * len(X_batch)\n",
    "    val_loss /= len(val_loader.dataset)\n",
    "    val_losses.append(val_loss)\n",
    "    \n",
    "    if val_loss == min(val_losses):\n",
    "        torch.save(model.state_dict(), 'saving.pth')\n",
    "\n",
    "    print(f'Epoch: {epoch+1}/{25},\\nTraining Loss:\\t {train_loss:0.20f},\\nValidation Loss: {val_loss:0.20f}')\n",
    "    print(\"#####################################################################################################\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "OIFKhbNweZzc"
   },
   "source": [
    "### Visualize Loss plot\n",
    "\n",
    "Using the arrays that you have (from previous task in the above section), visualize Loss plot (train and validation together)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 452
    },
    "id": "-uTbAB5YemMp",
    "outputId": "c788e62e-3784-4882-c780-4fd58135eba1"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12, 10))\n",
    "plt.plot(train_losses, label='train loss')\n",
    "plt.plot(val_losses, label='val loss')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "MRo2s0wMakLe"
   },
   "source": [
    "## Evaluation\n",
    "\n",
    "Test your trained model (using the Test Dataloader that you have)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "6TplSeYPakTT",
    "outputId": "fee4dbf4-6a55-423f-d051-262cbbac427b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 90.95432931549264\n"
     ]
    }
   ],
   "source": [
    "#TODO\n",
    "model.load_state_dict(torch.load('saving.pth'))\n",
    "model.eval()\n",
    "test_loss = 0\n",
    "y_true = []\n",
    "y_pred = []\n",
    "with torch.no_grad():\n",
    "    for X_batch, y_batch in test_loader:\n",
    "        y_true.extend(y_batch.cpu().numpy())\n",
    "        y_pred.extend(model(X_batch).cpu().numpy())\n",
    "        loss = criterion(model(X_batch), y_batch)\n",
    "        test_loss += loss.item() * len(X_batch)\n",
    "test_loss /= len(test_loader.dataset)\n",
    "\n",
    "y_true = np.array(y_true).reshape(-1)\n",
    "y_pred = np.array(y_pred).reshape(-1)\n",
    "y_pred = np.where(y_pred > 0.5, 1, 0)\n",
    "\n",
    "accuracy = (y_true == y_pred).mean()\n",
    "\n",
    "print('Accuracy:', accuracy * 100)"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "gpuType": "T4",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}