WideDeepCrossNetworks.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "v93Zaj2TtGDf"
      },
      "source": [
        "# Week 6 - Exercise 1\n",
        "\n",
        "Author: Khushee Kapoor\n",
        "\n",
        "Registration Number: 200968052\n",
        "\n",
        "Last Updated: 18/9/22"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WA2FYA3PtQUb"
      },
      "source": [
        "## Setting Up"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1fsVXjTGu3yx"
      },
      "source": [
        "**Meta-Data** of the dataset: (Name / Data Type / Measurement / Description)\n",
        "\n",
        "- Elevation / quantitative /meters / Elevation in meters\n",
        "- Aspect / quantitative / azimuth / Aspect in degrees azimutth\n",
        "- Slope / quantitative / degrees / Slope in degrees\n",
        "- Horizontal_Distance_To_Hydrology / quantitative / meters / Horz Dist to nearest surface water features\n",
        "- Vertical_Distance_To_Hydrology / quantitative / meters / Vert Dist to nearest surface water features\n",
        "- Horizontal_Distance_To_Roadways / quantitative / meters / Horz Dist to nearest roadway\n",
        "- Hillshade_9am / quantitative / 0 to 255 index / Hillshade index at 9am, summer solstice\n",
        "- Hillshade_Noon / quantitative / 0 to 255 index / Hillshade index at noon, summer soltice\n",
        "- Hillshade_3pm / quantitative / 0 to 255 index / Hillshade index at 3pm, summer solstice\n",
        "- Horizontal_Distance_To_Fire_Points / quantitative / meters / Horz Dist to nearest wildfire ignition points\n",
        "- Wilderness_Area (4 binary columns) / qualitative / 0 (absence) or 1 (presence) / Wilderness area designation\n",
        "- Soil_Type (40 binary columns) / qualitative / 0 (absence) or 1 (presence) / Soil Type designation\n",
        "- Cover_Type (7 types) / integer / 1 to 7 / Forest Cover Type designation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d2wREKVotQsC"
      },
      "source": [
        "To set up, we import the following libraries:\n",
        "\n",
        "- math: for mathematical functions\n",
        "- numpy: for data manipulation\n",
        "- pandas: for data manipulation\n",
        "- matplotlib: for data visualization\n",
        "- seaborn: for data visualization\n",
        "- tensorflow: to build neural networks"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "FDtFBjeTKLW-"
      },
      "outputs": [],
      "source": [
        "# importing libraries\n",
        "import math\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "from matplotlib import pyplot as plt\n",
        "import seaborn as sns\n",
        "import tensorflow as tf"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lVSidEPKtZyW"
      },
      "source": [
        "Next, we read the dataset from the link given using the read_csv() function from the Pandas library."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "c-w5K4bUKuGQ"
      },
      "outputs": [],
      "source": [
        "# reading the dataset\n",
        "data_url = ('https://archive.ics.uci.edu/ml/machine-learning-databases/covtype/covtype.data.gz')\n",
        "raw_data = pd.read_csv(data_url, header=None)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o2IR8sb8tiF1"
      },
      "source": [
        "To check if we have read the data correctly, we use the head() funtion from the Pandas library to view the first few rows."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        },
        "id": "U2t-_8HRSjVZ",
        "outputId": "0c285995-aec3-47c5-93b4-00cd0876679d"
      },
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-0ed9da02-757a-43ab-990c-c6dcd6722eb1\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <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",
              "      <th>5</th>\n",
              "      <th>6</th>\n",
              "      <th>7</th>\n",
              "      <th>8</th>\n",
              "      <th>9</th>\n",
              "      <th>...</th>\n",
              "      <th>45</th>\n",
              "      <th>46</th>\n",
              "      <th>47</th>\n",
              "      <th>48</th>\n",
              "      <th>49</th>\n",
              "      <th>50</th>\n",
              "      <th>51</th>\n",
              "      <th>52</th>\n",
              "      <th>53</th>\n",
              "      <th>54</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>2596</td>\n",
              "      <td>51</td>\n",
              "      <td>3</td>\n",
              "      <td>258</td>\n",
              "      <td>0</td>\n",
              "      <td>510</td>\n",
              "      <td>221</td>\n",
              "      <td>232</td>\n",
              "      <td>148</td>\n",
              "      <td>6279</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2590</td>\n",
              "      <td>56</td>\n",
              "      <td>2</td>\n",
              "      <td>212</td>\n",
              "      <td>-6</td>\n",
              "      <td>390</td>\n",
              "      <td>220</td>\n",
              "      <td>235</td>\n",
              "      <td>151</td>\n",
              "      <td>6225</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2804</td>\n",
              "      <td>139</td>\n",
              "      <td>9</td>\n",
              "      <td>268</td>\n",
              "      <td>65</td>\n",
              "      <td>3180</td>\n",
              "      <td>234</td>\n",
              "      <td>238</td>\n",
              "      <td>135</td>\n",
              "      <td>6121</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>2785</td>\n",
              "      <td>155</td>\n",
              "      <td>18</td>\n",
              "      <td>242</td>\n",
              "      <td>118</td>\n",
              "      <td>3090</td>\n",
              "      <td>238</td>\n",
              "      <td>238</td>\n",
              "      <td>122</td>\n",
              "      <td>6211</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>2595</td>\n",
              "      <td>45</td>\n",
              "      <td>2</td>\n",
              "      <td>153</td>\n",
              "      <td>-1</td>\n",
              "      <td>391</td>\n",
              "      <td>220</td>\n",
              "      <td>234</td>\n",
              "      <td>150</td>\n",
              "      <td>6172</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>5 rows × 55 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-0ed9da02-757a-43ab-990c-c6dcd6722eb1')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-0ed9da02-757a-43ab-990c-c6dcd6722eb1 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-0ed9da02-757a-43ab-990c-c6dcd6722eb1');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ],
            "text/plain": [
              "     0    1   2    3    4     5    6    7    8     9   ...  45  46  47  48  \\\n",
              "0  2596   51   3  258    0   510  221  232  148  6279  ...   0   0   0   0   \n",
              "1  2590   56   2  212   -6   390  220  235  151  6225  ...   0   0   0   0   \n",
              "2  2804  139   9  268   65  3180  234  238  135  6121  ...   0   0   0   0   \n",
              "3  2785  155  18  242  118  3090  238  238  122  6211  ...   0   0   0   0   \n",
              "4  2595   45   2  153   -1   391  220  234  150  6172  ...   0   0   0   0   \n",
              "\n",
              "   49  50  51  52  53  54  \n",
              "0   0   0   0   0   0   5  \n",
              "1   0   0   0   0   0   5  \n",
              "2   0   0   0   0   0   2  \n",
              "3   0   0   0   0   0   2  \n",
              "4   0   0   0   0   0   5  \n",
              "\n",
              "[5 rows x 55 columns]"
            ]
          },
          "execution_count": 3,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "# viewing the first few rows\n",
        "raw_data.head()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "djTeVCVTtpNZ"
      },
      "source": [
        "The two categorical features in the dataset are binary-encoded. We will convert this dataset representation to the typical representation, where each categorical feature is represented as a single integer value."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "jlOoE4cCTmpp"
      },
      "outputs": [],
      "source": [
        "# converting the categorical features to be represented as a single integer value\n",
        "soil_type_values = [f\"soil_type_{idx+1}\" for idx in range(40)]\n",
        "wilderness_area_values = [f\"area_type_{idx+1}\" for idx in range(4)]\n",
        "soil_type = raw_data.loc[:, 14:53].apply(lambda x: soil_type_values[0::1][x.to_numpy().nonzero()[0][0]], axis=1)\n",
        "wilderness_area = raw_data.loc[:, 10:13].apply(lambda x: wilderness_area_values[0::1][x.to_numpy().nonzero()[0][0]], axis=1)\n",
        "data = pd.concat([raw_data.loc[:, 0:9], wilderness_area, soil_type, raw_data.loc[:, 54]], axis=1, ignore_index=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FdVj3VfLt_zc"
      },
      "source": [
        "To make the data more readable, we add in the column labels from the metadata of the dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "id": "LVS44YVsh2yD"
      },
      "outputs": [],
      "source": [
        "# adding column labels\n",
        "column_labels = [\"Elevation\", \"Aspect\", \"Slope\", \"Horizontal_Distance_To_Hydrology\", \"Vertical_Distance_To_Hydrology\", \"Horizontal_Distance_To_Roadways\", \"Hillshade_9am\",\n",
        "                  \"Hillshade_Noon\", \"Hillshade_3pm\", \"Horizontal_Distance_To_Fire_Points\", \"Wilderness_Area\", \"Soil_Type\", \"Cover_Type\"]\n",
        "data.columns = column_labels"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NmZBeHs1vRkX"
      },
      "source": [
        "For ease in computation, we convert the target labels from 1-7 to be 0-6 by subtracting one from all the values."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "id": "Z6jfX7kPnMN-"
      },
      "outputs": [],
      "source": [
        "# converting the target labels\n",
        "data[\"Cover_Type\"] = data[\"Cover_Type\"] - 1"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ft4dxOhSvhVp"
      },
      "source": [
        "Next, we split the dataset into training and testing sets. We allocate 85% to training and 15% to the testing set."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "id": "wORy5yj3iZXj"
      },
      "outputs": [],
      "source": [
        "train_splits = []\n",
        "test_splits = []\n",
        "\n",
        "# splitting the dataset\n",
        "for _, group_data in data.groupby(\"Cover_Type\"):\n",
        "    random_selection = np.random.rand(len(group_data.index)) <= 0.85\n",
        "    train_splits.append(group_data[random_selection])\n",
        "    test_splits.append(group_data[~random_selection])\n",
        "\n",
        "train_data = pd.concat(train_splits).sample(frac=1).reset_index(drop=True)\n",
        "test_data = pd.concat(test_splits).sample(frac=1).reset_index(drop=True)\n",
        "\n",
        "# saving the training and testing sets as csv files\n",
        "train_data_file = \"train_data.csv\"\n",
        "test_data_file = \"test_data.csv\"\n",
        "train_data.to_csv(train_data_file, index=False)\n",
        "test_data.to_csv(test_data_file, index=False)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Jen8YjNfx5Vq"
      },
      "source": [
        "To make the meta-data more readable, we define the metadata of the dataset that will be useful for reading and parsing the data into input features, and encoding the input features with respect to their types."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "sg7IW0F4idqj"
      },
      "outputs": [],
      "source": [
        "# defining the target feature names and labels\n",
        "target_feature_name = \"Cover_Type\"\n",
        "target_feature_labels = [\"0\", \"1\", \"2\", \"3\", \"4\", \"5\", \"6\"]\n",
        "\n",
        "# defining the numeric feature names\n",
        "numeric_feature_names = [\"Aspect\", \"Elevation\", \"Hillshade_3pm\", \"Hillshade_9am\", \"Hillshade_Noon\", \"Horizontal_Distance_To_Fire_Points\", \"Horizontal_Distance_To_Hydrology\", \n",
        "                         \"Horizontal_Distance_To_Roadways\", \"Slope\", \"Vertical_Distance_To_Hydrology\"]\n",
        "\n",
        "# defining the categorical feature names\n",
        "categorical_features_vocabulary = {\n",
        "    \"Soil_Type\": list(data[\"Soil_Type\"].unique()),\n",
        "    \"Wilderness_Area\": list(data[\"Wilderness_Area\"].unique()),\n",
        "}\n",
        "categorical_feature_names = list(categorical_features_vocabulary.keys())\n",
        "\n",
        "# combining the numeric and categorical features\n",
        "feature_names = numeric_feature_names + categorical_feature_names\n",
        "\n",
        "# defining the column defaults\n",
        "column_defaults = [\n",
        "    [0] if feature_name in numeric_feature_names + [target_feature_name] else [\"NA\"]\n",
        "    for feature_name in column_labels\n",
        "]\n",
        "\n",
        "# obtaining the number of classes\n",
        "num_classes = len(target_feature_labels)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lku7BaBLyUqH"
      },
      "source": [
        "Next, we define the hyperparameters for training the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "id": "HSjyqXWzjciM"
      },
      "outputs": [],
      "source": [
        "# defining the hyperparameters\n",
        "learning_rate = 0.001\n",
        "dropout_rate = 0.1\n",
        "batch_size = 265\n",
        "num_epochs = 50\n",
        "hidden_units = [32, 32]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HO4mRrqlydra"
      },
      "source": [
        "We then prepare the training and testing sets by combining them with the above defined meta data."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "_xUxqCQWrltm"
      },
      "outputs": [],
      "source": [
        "# preparing the training and testing sets\n",
        "train_dataset = tf.data.experimental.make_csv_dataset(train_data_file, batch_size=batch_size, column_names=column_labels, column_defaults=column_defaults,\n",
        "        label_name=target_feature_name, num_epochs=1, header=True, shuffle=False).cache()\n",
        "test_dataset = tf.data.experimental.make_csv_dataset(test_data_file, batch_size=batch_size, column_names=column_labels, column_defaults=column_defaults,\n",
        "        label_name=target_feature_name, num_epochs=1, header=True, shuffle=False).cache()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "24Vb2Sfiyz5a"
      },
      "source": [
        "## Feed-Forward Network"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hY6Tj9eP5g73"
      },
      "source": [
        "To begin, we define the inputs for the models as a dictionary, where the key is the feature name, and the value is a keras.layers.Input tensor with the corresponding feature shape and data type."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "id": "0axfPIRoQ3J8"
      },
      "outputs": [],
      "source": [
        "# defining the inputs\n",
        "inputs = {}\n",
        "for feature_name in feature_names:\n",
        "    if feature_name in numeric_feature_names:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.float32)\n",
        "    else:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.string)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w1ym9_VZ5pM7"
      },
      "source": [
        "Next, we create a sparse representation of our input features wherein the categorical features are encoded with one-hot encoding."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3Rdy4TCiQ2ca",
        "outputId": "40971a92-47de-4de7-91be-3ff9a84e5bb7"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/numpy/core/numeric.py:2446: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison\n",
            "  return bool(asarray(a1 == a2).all())\n"
          ]
        }
      ],
      "source": [
        "# creating a sparse representation\n",
        "encoded_features = []\n",
        "for feature_name in inputs:\n",
        "    if feature_name in categorical_feature_names:\n",
        "        vocabulary = categorical_features_vocabulary[feature_name]\n",
        "        lookup = tf.keras.layers.StringLookup(vocabulary=vocabulary, mask_token=None, num_oov_indices=0, output_mode=\"binary\")\n",
        "        encoded_feature = lookup(tf.expand_dims(inputs[feature_name], -1))\n",
        "    else:\n",
        "        encoded_feature = tf.expand_dims(inputs[feature_name], -1)\n",
        "    encoded_features.append(encoded_feature)\n",
        "\n",
        "features = tf.keras.layers.concatenate(encoded_features)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cm2DwCQuL-tA"
      },
      "source": [
        "After that, we create two Dense layers with the given specifications. Then we create an Output layer and link the layers with the input and output."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "pZPpUlteRdPQ"
      },
      "outputs": [],
      "source": [
        "# creating two Dense layers\n",
        "for units in hidden_units: \n",
        "  features = tf.keras.layers.Dense(units)(features)\n",
        "  features = tf.keras.layers.BatchNormalization()(features)\n",
        "  features = tf.keras.layers.ReLU()(features)\n",
        "  features = tf.keras.layers.Dropout(dropout_rate)(features)\n",
        "\n",
        "# creating an output layer\n",
        "outputs = tf.keras.layers.Dense(units=num_classes, activation=\"softmax\")(features)\n",
        "\n",
        "# linking the layers with input and output\n",
        "basic_model = tf.keras.Model(inputs=inputs, outputs=outputs)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dWslLBryMxsO"
      },
      "source": [
        "Next compile the model and set the optimizer to be 'adam', loss to be SparseCategoricalCrossentropy and metric to be SparseCategoricalAccuracy. After that, we train the model on the training data with number of epochs set to the predefined value."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9sJLaliEkQW6",
        "outputId": "7cdded70-07e5-41ae-ee05-d9a1761a2d19"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1864/1864 [==============================] - 16s 8ms/step - loss: 0.7735 - sparse_categorical_accuracy: 0.6781\n",
            "Epoch 2/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.6645 - sparse_categorical_accuracy: 0.7138\n",
            "Epoch 3/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.6387 - sparse_categorical_accuracy: 0.7247\n",
            "Epoch 4/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.6181 - sparse_categorical_accuracy: 0.7359\n",
            "Epoch 5/50\n",
            "1864/1864 [==============================] - 11s 6ms/step - loss: 0.6058 - sparse_categorical_accuracy: 0.7405\n",
            "Epoch 6/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5958 - sparse_categorical_accuracy: 0.7446\n",
            "Epoch 7/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5896 - sparse_categorical_accuracy: 0.7476\n",
            "Epoch 8/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5834 - sparse_categorical_accuracy: 0.7488\n",
            "Epoch 9/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5789 - sparse_categorical_accuracy: 0.7514\n",
            "Epoch 10/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5755 - sparse_categorical_accuracy: 0.7529\n",
            "Epoch 11/50\n",
            "1864/1864 [==============================] - 9s 5ms/step - loss: 0.5733 - sparse_categorical_accuracy: 0.7532\n",
            "Epoch 12/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5697 - sparse_categorical_accuracy: 0.7553\n",
            "Epoch 13/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5673 - sparse_categorical_accuracy: 0.7558\n",
            "Epoch 14/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5660 - sparse_categorical_accuracy: 0.7561\n",
            "Epoch 15/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5646 - sparse_categorical_accuracy: 0.7572\n",
            "Epoch 16/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5626 - sparse_categorical_accuracy: 0.7576\n",
            "Epoch 17/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5617 - sparse_categorical_accuracy: 0.7580\n",
            "Epoch 18/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5595 - sparse_categorical_accuracy: 0.7591\n",
            "Epoch 19/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5576 - sparse_categorical_accuracy: 0.7600\n",
            "Epoch 20/50\n",
            "1864/1864 [==============================] - 9s 5ms/step - loss: 0.5575 - sparse_categorical_accuracy: 0.7601\n",
            "Epoch 21/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5555 - sparse_categorical_accuracy: 0.7612\n",
            "Epoch 22/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5545 - sparse_categorical_accuracy: 0.7612\n",
            "Epoch 23/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5527 - sparse_categorical_accuracy: 0.7628\n",
            "Epoch 24/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5513 - sparse_categorical_accuracy: 0.7630\n",
            "Epoch 25/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5512 - sparse_categorical_accuracy: 0.7630\n",
            "Epoch 26/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5503 - sparse_categorical_accuracy: 0.7635\n",
            "Epoch 27/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5482 - sparse_categorical_accuracy: 0.7645\n",
            "Epoch 28/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5468 - sparse_categorical_accuracy: 0.7648\n",
            "Epoch 29/50\n",
            "1864/1864 [==============================] - 9s 5ms/step - loss: 0.5462 - sparse_categorical_accuracy: 0.7659\n",
            "Epoch 30/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5449 - sparse_categorical_accuracy: 0.7662\n",
            "Epoch 31/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5435 - sparse_categorical_accuracy: 0.7667\n",
            "Epoch 32/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5430 - sparse_categorical_accuracy: 0.7673\n",
            "Epoch 33/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5426 - sparse_categorical_accuracy: 0.7670\n",
            "Epoch 34/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5418 - sparse_categorical_accuracy: 0.7681\n",
            "Epoch 35/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5405 - sparse_categorical_accuracy: 0.7680\n",
            "Epoch 36/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5410 - sparse_categorical_accuracy: 0.7681\n",
            "Epoch 37/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5389 - sparse_categorical_accuracy: 0.7685\n",
            "Epoch 38/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5390 - sparse_categorical_accuracy: 0.7690\n",
            "Epoch 39/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5375 - sparse_categorical_accuracy: 0.7692\n",
            "Epoch 40/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5376 - sparse_categorical_accuracy: 0.7692\n",
            "Epoch 41/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5370 - sparse_categorical_accuracy: 0.7692\n",
            "Epoch 42/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5362 - sparse_categorical_accuracy: 0.7698\n",
            "Epoch 43/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5354 - sparse_categorical_accuracy: 0.7699\n",
            "Epoch 44/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5354 - sparse_categorical_accuracy: 0.7704\n",
            "Epoch 45/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5347 - sparse_categorical_accuracy: 0.7709\n",
            "Epoch 46/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5340 - sparse_categorical_accuracy: 0.7712\n",
            "Epoch 47/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5338 - sparse_categorical_accuracy: 0.7706\n",
            "Epoch 48/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5337 - sparse_categorical_accuracy: 0.7714\n",
            "Epoch 49/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5333 - sparse_categorical_accuracy: 0.7709\n",
            "Epoch 50/50\n",
            "1864/1864 [==============================] - 10s 5ms/step - loss: 0.5330 - sparse_categorical_accuracy: 0.7719\n"
          ]
        }
      ],
      "source": [
        "# compiling the model\n",
        "basic_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate), loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])\n",
        "\n",
        "# training the model\n",
        "basic_history = basic_model.fit(train_dataset, epochs=num_epochs, batch_size=batch_size)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "juRBJaWPNDwk"
      },
      "source": [
        "## Wide & Deep Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wL4GJZuGNI5u"
      },
      "source": [
        "To begin, we define the inputs for the models as a dictionary, where the key is the feature name, and the value is a keras.layers.Input tensor with the corresponding feature shape and data type."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "id": "fuXs-ixbp75P"
      },
      "outputs": [],
      "source": [
        "# defining the inputs\n",
        "inputs = {}\n",
        "for feature_name in feature_names:\n",
        "    if feature_name in numeric_feature_names:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.float32)\n",
        "    else:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.string)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x4l9rHXGNO3V"
      },
      "source": [
        "Next, we create a sparse representation of our input features wherein the categorical features are encoded with one-hot encoding for the wide part of the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "H7nBk-dwp75Q"
      },
      "outputs": [],
      "source": [
        "# creating a sparse representation for the wide part\n",
        "encoded_features = []\n",
        "for feature_name in inputs:\n",
        "    if feature_name in categorical_feature_names:\n",
        "        vocabulary = categorical_features_vocabulary[feature_name]\n",
        "        lookup = tf.keras.layers.StringLookup(vocabulary=vocabulary, mask_token=None, num_oov_indices=0, output_mode=\"binary\")\n",
        "        encoded_feature = lookup(tf.expand_dims(inputs[feature_name], -1))\n",
        "    else:\n",
        "        encoded_feature = tf.expand_dims(inputs[feature_name], -1)\n",
        "    encoded_features.append(encoded_feature)\n",
        "\n",
        "wide = tf.keras.layers.concatenate(encoded_features)\n",
        "wide = tf.keras.layers.BatchNormalization()(wide)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pUH12tIbNaoF"
      },
      "source": [
        "After that, we create a dense represenation of our input features wherein the categorical features are encoded with low-dimensional embeddings for the deep part of the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "id": "Q6AaX3dAp9DG"
      },
      "outputs": [],
      "source": [
        "# creating a dense representation for the deep part\n",
        "encoded_features = []\n",
        "for feature_name in inputs:\n",
        "    if feature_name in categorical_feature_names:\n",
        "        vocabulary = categorical_features_vocabulary[feature_name]\n",
        "        lookup = tf.keras.layers.StringLookup(vocabulary=vocabulary, mask_token=None, num_oov_indices=0, output_mode=\"int\")\n",
        "        encoded_feature = lookup(inputs[feature_name])\n",
        "        embedding_dims = int(math.sqrt(len(vocabulary)))\n",
        "        embedding = tf.keras.layers.Embedding(input_dim=len(vocabulary), output_dim=embedding_dims)\n",
        "        encoded_feature = embedding(encoded_feature)\n",
        "    else:\n",
        "        encoded_feature = tf.expand_dims(inputs[feature_name], -1)\n",
        "\n",
        "    encoded_features.append(encoded_feature)\n",
        "\n",
        "deep = tf.keras.layers.concatenate(encoded_features)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jWEEN4Y4NvyR"
      },
      "source": [
        "Now, we create a deep network having two layers and predefined specifications, and merge it with the wide part. Then we create an Output layer and link the layers with the input and output."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "metadata": {
        "id": "U2W0k3MIq1Ji"
      },
      "outputs": [],
      "source": [
        "# creating a deep network\n",
        "for units in hidden_units:\n",
        "  deep = tf.keras.layers.Dense(units)(deep)\n",
        "  deep = tf.keras.layers.BatchNormalization()(deep)\n",
        "  deep = tf.keras.layers.ReLU()(deep)\n",
        "  deep = tf.keras.layers.Dropout(dropout_rate)(deep)\n",
        "\n",
        "# merging the wide and deep network\n",
        "merged = tf.keras.layers.concatenate([wide, deep])\n",
        "\n",
        "# creating an output layer\n",
        "outputs = tf.keras.layers.Dense(units=num_classes, activation=\"softmax\")(merged)\n",
        "\n",
        "# linking the layers with the input and output\n",
        "wide_deep_model = tf.keras.Model(inputs=inputs, outputs=outputs)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W2Hb8uiDOM0O"
      },
      "source": [
        "Next compile the model and set the optimizer to be 'adam', loss to be SparseCategoricalCrossentropy and metric to be SparseCategoricalAccuracy. After that, we train the model on the training data with number of epochs set to the predefined value."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 19,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bN-BvZLNq5Lu",
        "outputId": "f0cbeef8-ab51-42da-f8d0-0e555506aa9b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1864/1864 [==============================] - 15s 7ms/step - loss: 0.7041 - sparse_categorical_accuracy: 0.7061\n",
            "Epoch 2/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.6042 - sparse_categorical_accuracy: 0.7383\n",
            "Epoch 3/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5887 - sparse_categorical_accuracy: 0.7438\n",
            "Epoch 4/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5761 - sparse_categorical_accuracy: 0.7499\n",
            "Epoch 5/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5662 - sparse_categorical_accuracy: 0.7545\n",
            "Epoch 6/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5581 - sparse_categorical_accuracy: 0.7591\n",
            "Epoch 7/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5525 - sparse_categorical_accuracy: 0.7611\n",
            "Epoch 8/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5467 - sparse_categorical_accuracy: 0.7639\n",
            "Epoch 9/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5429 - sparse_categorical_accuracy: 0.7663\n",
            "Epoch 10/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5391 - sparse_categorical_accuracy: 0.7681\n",
            "Epoch 11/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5356 - sparse_categorical_accuracy: 0.7692\n",
            "Epoch 12/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5332 - sparse_categorical_accuracy: 0.7700\n",
            "Epoch 13/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5302 - sparse_categorical_accuracy: 0.7715\n",
            "Epoch 14/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5275 - sparse_categorical_accuracy: 0.7729\n",
            "Epoch 15/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5253 - sparse_categorical_accuracy: 0.7738\n",
            "Epoch 16/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5231 - sparse_categorical_accuracy: 0.7751\n",
            "Epoch 17/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5218 - sparse_categorical_accuracy: 0.7756\n",
            "Epoch 18/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5201 - sparse_categorical_accuracy: 0.7758\n",
            "Epoch 19/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5183 - sparse_categorical_accuracy: 0.7776\n",
            "Epoch 20/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5164 - sparse_categorical_accuracy: 0.7784\n",
            "Epoch 21/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5149 - sparse_categorical_accuracy: 0.7791\n",
            "Epoch 22/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5138 - sparse_categorical_accuracy: 0.7797\n",
            "Epoch 23/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5129 - sparse_categorical_accuracy: 0.7802\n",
            "Epoch 24/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5111 - sparse_categorical_accuracy: 0.7816\n",
            "Epoch 25/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5100 - sparse_categorical_accuracy: 0.7821\n",
            "Epoch 26/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5093 - sparse_categorical_accuracy: 0.7820\n",
            "Epoch 27/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5079 - sparse_categorical_accuracy: 0.7837\n",
            "Epoch 28/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5071 - sparse_categorical_accuracy: 0.7839\n",
            "Epoch 29/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5071 - sparse_categorical_accuracy: 0.7838\n",
            "Epoch 30/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5053 - sparse_categorical_accuracy: 0.7845\n",
            "Epoch 31/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5043 - sparse_categorical_accuracy: 0.7853\n",
            "Epoch 32/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5044 - sparse_categorical_accuracy: 0.7852\n",
            "Epoch 33/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5033 - sparse_categorical_accuracy: 0.7858\n",
            "Epoch 34/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5014 - sparse_categorical_accuracy: 0.7867\n",
            "Epoch 35/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.5014 - sparse_categorical_accuracy: 0.7868\n",
            "Epoch 36/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5014 - sparse_categorical_accuracy: 0.7865\n",
            "Epoch 37/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5008 - sparse_categorical_accuracy: 0.7865\n",
            "Epoch 38/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4996 - sparse_categorical_accuracy: 0.7875\n",
            "Epoch 39/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4996 - sparse_categorical_accuracy: 0.7869\n",
            "Epoch 40/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4988 - sparse_categorical_accuracy: 0.7884\n",
            "Epoch 41/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4985 - sparse_categorical_accuracy: 0.7881\n",
            "Epoch 42/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4985 - sparse_categorical_accuracy: 0.7887\n",
            "Epoch 43/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4975 - sparse_categorical_accuracy: 0.7890\n",
            "Epoch 44/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4976 - sparse_categorical_accuracy: 0.7885\n",
            "Epoch 45/50\n",
            "1864/1864 [==============================] - 14s 7ms/step - loss: 0.4975 - sparse_categorical_accuracy: 0.7881\n",
            "Epoch 46/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4968 - sparse_categorical_accuracy: 0.7888\n",
            "Epoch 47/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.4966 - sparse_categorical_accuracy: 0.7888\n",
            "Epoch 48/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.4957 - sparse_categorical_accuracy: 0.7892\n",
            "Epoch 49/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.4949 - sparse_categorical_accuracy: 0.7897\n",
            "Epoch 50/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.4948 - sparse_categorical_accuracy: 0.7893\n"
          ]
        }
      ],
      "source": [
        "# compiling the model\n",
        "wide_deep_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate), loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])\n",
        "\n",
        "# training the model\n",
        "wide_deep_history = wide_deep_model.fit(train_dataset, epochs=num_epochs, batch_size=batch_size)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UsutJur7OqTj"
      },
      "source": [
        "## Deep & Cross Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aCqwzggHPC_K"
      },
      "source": [
        "To begin, we define the inputs for the models as a dictionary, where the key is the feature name, and the value is a keras.layers.Input tensor with the corresponding feature shape and data type."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {
        "id": "0E-DDXGYr65m"
      },
      "outputs": [],
      "source": [
        "# defining the inputs\n",
        "inputs = {}\n",
        "for feature_name in feature_names:\n",
        "    if feature_name in numeric_feature_names:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.float32)\n",
        "    else:\n",
        "        inputs[feature_name] = tf.keras.layers.Input(name=feature_name, shape=(), dtype=tf.string)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D8M9xywfPWyY"
      },
      "source": [
        "After that, we create a dense represenation of our input features wherein the categorical features are encoded with low-dimensional embeddings for the deep part of the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 21,
      "metadata": {
        "id": "bY3Cp2F9r_Om"
      },
      "outputs": [],
      "source": [
        "# creating a dense representation for the deep part\n",
        "encoded_features = []\n",
        "for feature_name in inputs:\n",
        "    if feature_name in categorical_feature_names:\n",
        "        vocabulary = categorical_features_vocabulary[feature_name]\n",
        "        lookup = tf.keras.layers.StringLookup(vocabulary=vocabulary, mask_token=None, num_oov_indices=0, output_mode=\"int\")\n",
        "        encoded_feature = lookup(inputs[feature_name])\n",
        "        embedding_dims = int(math.sqrt(len(vocabulary)))\n",
        "        embedding = tf.keras.layers.Embedding(input_dim=len(vocabulary), output_dim=embedding_dims)\n",
        "        encoded_feature = embedding(encoded_feature)\n",
        "    else:\n",
        "        encoded_feature = tf.expand_dims(inputs[feature_name], -1)\n",
        "\n",
        "    encoded_features.append(encoded_feature)\n",
        "\n",
        "x0 = tf.keras.layers.concatenate(encoded_features)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "G5R_570kPd9J"
      },
      "source": [
        "Now, we create a cross part in a way that the explicit feature crossing are applied in an efficient way, where the degree of cross features grows with layer depth. Next, we create a deep network having two layers and predefined specifications, and merge it with the cross part. Then we create an Output layer and link the layers with the input and output."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 22,
      "metadata": {
        "id": "DCbUIw_asF6e"
      },
      "outputs": [],
      "source": [
        "# creating the cross part\n",
        "cross = x0\n",
        "for _ in hidden_units:\n",
        "    units = cross.shape[-1]\n",
        "    x = tf.keras.layers.Dense(units)(cross)\n",
        "    cross = x0 * x + cross\n",
        "cross = tf.keras.layers.BatchNormalization()(cross)\n",
        "\n",
        "deep = x0\n",
        "\n",
        "# creating a deep network\n",
        "for units in hidden_units:\n",
        "    deep = tf.keras.layers.Dense(units)(deep)\n",
        "    deep = tf.keras.layers.BatchNormalization()(deep)\n",
        "    deep = tf.keras.layers.ReLU()(deep)\n",
        "    deep = tf.keras.layers.Dropout(dropout_rate)(deep)\n",
        "\n",
        "# merging the cross and deep part\n",
        "merged = tf.keras.layers.concatenate([cross, deep])\n",
        "\n",
        "# creating an output layer\n",
        "outputs = tf.keras.layers.Dense(units=num_classes, activation=\"softmax\")(merged)\n",
        "\n",
        "# linking the layers with the inputs and outputs\n",
        "deep_cross_model = tf.keras.Model(inputs=inputs, outputs=outputs)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jeiB63d6QQ0o"
      },
      "source": [
        "Next compile the model and set the optimizer to be 'adam', loss to be SparseCategoricalCrossentropy and metric to be SparseCategoricalAccuracy. After that, we train the model on the training data with number of epochs set to the predefined value."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 23,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2GfWBwCgsW6H",
        "outputId": "8a7de29d-b163-495a-a5cd-990daceb5b38"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1864/1864 [==============================] - 13s 6ms/step - loss: 0.6995 - sparse_categorical_accuracy: 0.7101\n",
            "Epoch 2/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5919 - sparse_categorical_accuracy: 0.7439\n",
            "Epoch 3/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5757 - sparse_categorical_accuracy: 0.7493\n",
            "Epoch 4/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5644 - sparse_categorical_accuracy: 0.7536\n",
            "Epoch 5/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5568 - sparse_categorical_accuracy: 0.7562\n",
            "Epoch 6/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5507 - sparse_categorical_accuracy: 0.7596\n",
            "Epoch 7/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5462 - sparse_categorical_accuracy: 0.7621\n",
            "Epoch 8/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5421 - sparse_categorical_accuracy: 0.7631\n",
            "Epoch 9/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5386 - sparse_categorical_accuracy: 0.7652\n",
            "Epoch 10/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5360 - sparse_categorical_accuracy: 0.7661\n",
            "Epoch 11/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5336 - sparse_categorical_accuracy: 0.7676\n",
            "Epoch 12/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5316 - sparse_categorical_accuracy: 0.7679\n",
            "Epoch 13/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5298 - sparse_categorical_accuracy: 0.7688\n",
            "Epoch 14/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5282 - sparse_categorical_accuracy: 0.7697\n",
            "Epoch 15/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5264 - sparse_categorical_accuracy: 0.7707\n",
            "Epoch 16/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5246 - sparse_categorical_accuracy: 0.7713\n",
            "Epoch 17/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5228 - sparse_categorical_accuracy: 0.7720\n",
            "Epoch 18/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5211 - sparse_categorical_accuracy: 0.7731\n",
            "Epoch 19/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5198 - sparse_categorical_accuracy: 0.7739\n",
            "Epoch 20/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5180 - sparse_categorical_accuracy: 0.7746\n",
            "Epoch 21/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5164 - sparse_categorical_accuracy: 0.7752\n",
            "Epoch 22/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5151 - sparse_categorical_accuracy: 0.7763\n",
            "Epoch 23/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5138 - sparse_categorical_accuracy: 0.7761\n",
            "Epoch 24/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5127 - sparse_categorical_accuracy: 0.7764\n",
            "Epoch 25/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5114 - sparse_categorical_accuracy: 0.7769\n",
            "Epoch 26/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5102 - sparse_categorical_accuracy: 0.7779\n",
            "Epoch 27/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5088 - sparse_categorical_accuracy: 0.7784\n",
            "Epoch 28/50\n",
            "1864/1864 [==============================] - 14s 8ms/step - loss: 0.5081 - sparse_categorical_accuracy: 0.7793\n",
            "Epoch 29/50\n",
            "1864/1864 [==============================] - 14s 8ms/step - loss: 0.5069 - sparse_categorical_accuracy: 0.7798\n",
            "Epoch 30/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5051 - sparse_categorical_accuracy: 0.7801\n",
            "Epoch 31/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5043 - sparse_categorical_accuracy: 0.7808\n",
            "Epoch 32/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5032 - sparse_categorical_accuracy: 0.7820\n",
            "Epoch 33/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.5019 - sparse_categorical_accuracy: 0.7825\n",
            "Epoch 34/50\n",
            "1864/1864 [==============================] - 12s 7ms/step - loss: 0.5007 - sparse_categorical_accuracy: 0.7828\n",
            "Epoch 35/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4999 - sparse_categorical_accuracy: 0.7836\n",
            "Epoch 36/50\n",
            "1864/1864 [==============================] - 13s 7ms/step - loss: 0.4991 - sparse_categorical_accuracy: 0.7846\n",
            "Epoch 37/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4979 - sparse_categorical_accuracy: 0.7853\n",
            "Epoch 38/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4971 - sparse_categorical_accuracy: 0.7857\n",
            "Epoch 39/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4960 - sparse_categorical_accuracy: 0.7860\n",
            "Epoch 40/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4942 - sparse_categorical_accuracy: 0.7870\n",
            "Epoch 41/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4943 - sparse_categorical_accuracy: 0.7869\n",
            "Epoch 42/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4934 - sparse_categorical_accuracy: 0.7876\n",
            "Epoch 43/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4929 - sparse_categorical_accuracy: 0.7877\n",
            "Epoch 44/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4924 - sparse_categorical_accuracy: 0.7885\n",
            "Epoch 45/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4912 - sparse_categorical_accuracy: 0.7886\n",
            "Epoch 46/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4908 - sparse_categorical_accuracy: 0.7884\n",
            "Epoch 47/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4902 - sparse_categorical_accuracy: 0.7891\n",
            "Epoch 48/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4897 - sparse_categorical_accuracy: 0.7891\n",
            "Epoch 49/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4892 - sparse_categorical_accuracy: 0.7896\n",
            "Epoch 50/50\n",
            "1864/1864 [==============================] - 12s 6ms/step - loss: 0.4885 - sparse_categorical_accuracy: 0.7898\n"
          ]
        }
      ],
      "source": [
        "# compiling the model\n",
        "deep_cross_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate), loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])\n",
        "\n",
        "# training the model\n",
        "deep_cross_history = deep_cross_model.fit(train_dataset, epochs=num_epochs, batch_size=batch_size)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZltsY1SFQVFb"
      },
      "source": [
        "## Comparison"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-ntVqmelQWFE"
      },
      "source": [
        "To compare the loss of the models, we plot them using the history attribute in a single plot and differentiate between them by giving different colors and appropriate labels using functions from the matplotlib library."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 24,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 530
        },
        "id": "kDcWL4Ok0fsC",
        "outputId": "9148da47-6ff8-4d24-b58e-b4b1f39f096b"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7fdda0060e90>"
            ]
          },
          "execution_count": 24,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1080x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "# plotting the losses of the models\n",
        "epochs = range(1, 51)\n",
        "plt.figure(figsize=(15,8))\n",
        "plt.plot(epochs, basic_history.history['loss'], 'b', label='Basic')\n",
        "plt.plot(epochs, wide_deep_history.history['loss'], 'r', label='Wide and Deep')\n",
        "plt.plot(epochs, deep_cross_history.history['loss'], 'k', label='Deep and Cross')\n",
        "plt.title('Comparing Losses of Models')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.legend()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eb91DRDfRFdI"
      },
      "source": [
        "As we can see, the loss of all the three models decreases and then plateuas. However, there is a significant difference between the basic model and the wide & deep and deep & cross models, which have very less difference, with basic model having higher loss than the other two."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "h7NIuvFpRUpX"
      },
      "source": [
        "To compare the accuracy of the models, we plot them using the history attribute in a single plot and differentiate between them by giving different colors and appropriate labels using functions from the matplotlib library."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 25,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 530
        },
        "id": "ggAa2ZsF1UUU",
        "outputId": "30cf9861-ccc3-4830-bc25-38cacf027d37"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7fdd3a24b5d0>"
            ]
          },
          "execution_count": 25,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 1080x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "# plotting the accuracies of the models\n",
        "epochs = range(1, 51)\n",
        "plt.figure(figsize=(15,8))\n",
        "plt.plot(epochs, basic_history.history['sparse_categorical_accuracy'], 'b', label='Basic')\n",
        "plt.plot(epochs, wide_deep_history.history['sparse_categorical_accuracy'], 'r', label='Wide and Deep')\n",
        "plt.plot(epochs, deep_cross_history.history['sparse_categorical_accuracy'], 'k', label='Deep and Cross')\n",
        "plt.title('Comparing Accuracies of Models')\n",
        "plt.ylabel('Accuracies')\n",
        "plt.xlabel('Epochs')\n",
        "plt.legend()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AHLb00OIRetr"
      },
      "source": [
        "As we can see, the accuracy of all the three models increases and then plateuas. However, there is a significant difference between the basic model and the wide & deep and deep & cross models, which have very less difference, with basic model having lower accuracy than the other two."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Da_zu_I5RjVb"
      },
      "source": [
        "Next, we evaluate the models on the test set using the evaluate() function. First, we compare the loss."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 26,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DlKB2Fda1hyb",
        "outputId": "97991d36-8b19-4939-f44e-08ca37961f4b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Loss of Basic Model on Test Set: 0.65638\n",
            "Loss of Wide & Deep Model on Test Set: 0.44760\n",
            "Loss of Deep & Cross Model on Test Set: 0.46343\n"
          ]
        }
      ],
      "source": [
        "# comparing the loss\n",
        "print(str.format('Loss of Basic Model on Test Set: {:.5f}', basic_model.evaluate(test_dataset, verbose=0)[0]))\n",
        "print(str.format('Loss of Wide & Deep Model on Test Set: {:.5f}', wide_deep_model.evaluate(test_dataset, verbose=0)[0]))\n",
        "print(str.format('Loss of Deep & Cross Model on Test Set: {:.5f}', deep_cross_model.evaluate(test_dataset, verbose=0)[0]))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zflvauJdRuSl"
      },
      "source": [
        "As we can see, the basic model has a higher loss than the wide & deep and deep & cross models, which are almost the same with the former being slightly less."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ESyArgmKR3Gn"
      },
      "source": [
        "Next, we compare the accuracy."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 27,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HRRoz4125L6P",
        "outputId": "f0950f91-2d8f-4114-92a8-a4341011bc5f"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Accuracy of Basic Model on Test Set: 73.60%\n",
            "Accuracy of Wide & Deep Model on Test Set: 80.94%\n",
            "Accuracy of Deep & Cross Model on Test Set: 80.10%\n"
          ]
        }
      ],
      "source": [
        "# comparing the accuracy\n",
        "print(str.format('Accuracy of Basic Model on Test Set: {:.2f}%', basic_model.evaluate(test_dataset, verbose=0)[1]*100))\n",
        "print(str.format('Accuracy of Wide & Deep Model on Test Set: {:.2f}%', wide_deep_model.evaluate(test_dataset, verbose=0)[1]*100))\n",
        "print(str.format('Accuracy of Deep & Cross Model on Test Set: {:.2f}%', deep_cross_model.evaluate(test_dataset, verbose=0)[1]*100))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xiSL-cdpR7vK"
      },
      "source": [
        "As we can see, the basic model has a lower accuracy than the wide & deep and deep & cross models, which are almost the same with the former being slightly more."
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3.8.10 64-bit (windows store)",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.8.10"
    },
    "vscode": {
      "interpreter": {
        "hash": "42b00969bc5f493d740a4024214575325c132751ccb925f4cd3f44cb5cce2e23"
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}