Gordon_RamsAI_v2.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Gordon RamsAI 5.0",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyOki3rg/WU31vogL4SKGpS9",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/TanushGoel/Gordon-RamsAI/blob/master/Gordon_RamsAI_v2.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mLZSBLyzr9Jw"
      },
      "source": [
        "!unzip -q -o \"/content/recipes_webscraped.csv.zip\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 221
        },
        "id": "f3N9JoFLtr7M",
        "outputId": "d98f5dc1-2b04-4f49-b7dc-8b097baa9c17"
      },
      "source": [
        "import pandas as pd\n",
        "\n",
        "data = pd.read_csv(\"/content/recipes_webscraped.csv\")\n",
        "print(len(data))\n",
        "data.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "517129\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>name</th>\n",
              "      <th>ingredients</th>\n",
              "      <th>steps</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>buttermilk pie with gingersnap crumb crust</td>\n",
              "      <td>sugar margarine egg egg whites flour salt butt...</td>\n",
              "      <td>preheat oven to 350°f. make pie crust, using 8...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>boston cream pie</td>\n",
              "      <td>margarine cake flour powder salt sugar oil van...</td>\n",
              "      <td>beat egg whites until soft peaks form. gradual...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>betty crocker's southwestern guacamole dip</td>\n",
              "      <td>avocados medium tomatoes salt cloves garlic li...</td>\n",
              "      <td>peel and mash avocados until lumpy. add remain...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>calzone</td>\n",
              "      <td>parker house rolls tomato sauce onion basil le...</td>\n",
              "      <td>open rolls and let stand until they reach room...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>buttermilk ranch dressing</td>\n",
              "      <td>clove garlic salt buttermilk sour cream lime j...</td>\n",
              "      <td>mash together garlic and salt to form a paste....</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                         name  ...                                              steps\n",
              "0  buttermilk pie with gingersnap crumb crust  ...  preheat oven to 350°f. make pie crust, using 8...\n",
              "1                            boston cream pie  ...  beat egg whites until soft peaks form. gradual...\n",
              "2  betty crocker's southwestern guacamole dip  ...  peel and mash avocados until lumpy. add remain...\n",
              "3                                     calzone  ...  open rolls and let stand until they reach room...\n",
              "4                   buttermilk ranch dressing  ...  mash together garlic and salt to form a paste....\n",
              "\n",
              "[5 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 1
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 337
        },
        "id": "LebPh9gou8or",
        "outputId": "71e4ddf0-93f5-4ece-e5ac-001d7173d3d4"
      },
      "source": [
        "import numpy as np\n",
        "\n",
        "pd.DataFrame([len(text.split()) for text in data.name])[0].hist(bins=np.arange(start=0, stop=50, step=1), figsize=[25,5])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f004630b9e8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1800x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 337
        },
        "id": "Eo-sIcUovDPU",
        "outputId": "1c5acf74-a245-4dd1-9b3e-9978b25d6c8b"
      },
      "source": [
        "pd.DataFrame([len(str(text).split()) for text in data.ingredients])[0].hist(bins=np.arange(start=0, stop=100, step=1), figsize=[25,5])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f004615d898>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1800x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 337
        },
        "id": "eKOJfK0FvDb0",
        "outputId": "52e13ce1-c54d-41e9-f0c2-2b5cc8fe252a"
      },
      "source": [
        "pd.DataFrame([len(str(text).split()) for text in data.steps])[0].hist(bins=np.arange(start=0, stop=750, step=1), figsize=[25,5])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f0045bec940>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1800x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 221
        },
        "id": "Qxg_c8uLv7YK",
        "outputId": "250576df-f8e5-433b-e8fe-76c0540d3a04"
      },
      "source": [
        "max_name_len = 24\n",
        "max_ingreds_len = 64\n",
        "max_steps_len = 512\n",
        "\n",
        "upper_bound_steps = 640\n",
        "data = data[data['steps'].str.split().str.len().le(upper_bound_steps)]\n",
        "\n",
        "print(len(data))\n",
        "data.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "516531\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>name</th>\n",
              "      <th>ingredients</th>\n",
              "      <th>steps</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>buttermilk pie with gingersnap crumb crust</td>\n",
              "      <td>sugar margarine egg egg whites flour salt butt...</td>\n",
              "      <td>preheat oven to 350°f. make pie crust, using 8...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>boston cream pie</td>\n",
              "      <td>margarine cake flour powder salt sugar oil van...</td>\n",
              "      <td>beat egg whites until soft peaks form. gradual...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>betty crocker's southwestern guacamole dip</td>\n",
              "      <td>avocados medium tomatoes salt cloves garlic li...</td>\n",
              "      <td>peel and mash avocados until lumpy. add remain...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>calzone</td>\n",
              "      <td>parker house rolls tomato sauce onion basil le...</td>\n",
              "      <td>open rolls and let stand until they reach room...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>buttermilk ranch dressing</td>\n",
              "      <td>clove garlic salt buttermilk sour cream lime j...</td>\n",
              "      <td>mash together garlic and salt to form a paste....</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                         name  ...                                              steps\n",
              "0  buttermilk pie with gingersnap crumb crust  ...  preheat oven to 350°f. make pie crust, using 8...\n",
              "1                            boston cream pie  ...  beat egg whites until soft peaks form. gradual...\n",
              "2  betty crocker's southwestern guacamole dip  ...  peel and mash avocados until lumpy. add remain...\n",
              "3                                     calzone  ...  open rolls and let stand until they reach room...\n",
              "4                   buttermilk ranch dressing  ...  mash together garlic and salt to form a paste....\n",
              "\n",
              "[5 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VSLxQeQetywo",
        "outputId": "49ac7b8f-7635-4069-b3e8-d7d01b6b537e"
      },
      "source": [
        "!pip3 install -q transformers\n",
        "import transformers\n",
        "\n",
        "tokenizer = transformers.BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=True)\n",
        "elements = len(tokenizer.get_vocab())\n",
        "print(elements)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "30522\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5Qh16xdnvBJe"
      },
      "source": [
        "def tokenize_name(sentence):\n",
        "\n",
        "    return tokenizer.encode_plus(str(sentence), max_length=max_name_len,\n",
        "                                 truncation=True, padding='max_length',\n",
        "                                 add_special_tokens=True, return_token_type_ids=False, return_attention_mask=False)['input_ids']\n",
        "\n",
        "def tokenize_ingreds(sentence):\n",
        "\n",
        "    return tokenizer.encode_plus(str(sentence), max_length=max_ingreds_len,\n",
        "                                 truncation=True, padding='max_length',\n",
        "                                 add_special_tokens=True, return_token_type_ids=False, return_attention_mask=False)['input_ids']\n",
        "\n",
        "def tokenize_steps(sentence):\n",
        "\n",
        "    return tokenizer.encode_plus(str(sentence), max_length=max_steps_len,\n",
        "                                 truncation=True, padding='max_length',\n",
        "                                 add_special_tokens=True, return_token_type_ids=False, return_attention_mask=False)['input_ids']\n",
        "\n",
        "def decode(tokenized):\n",
        "\n",
        "    return tokenizer.decode(tokenized, skip_special_tokens=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RydL558myUKn"
      },
      "source": [
        "data.steps.iloc[0]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d1hpSyAAy2Oy"
      },
      "source": [
        "print(tokenize_steps(data.steps.iloc[0]))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "501GGOBbxbD3"
      },
      "source": [
        "decode(tokenize_steps(data.steps.iloc[0]))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cMIyjdeM7Ghk"
      },
      "source": [
        "def augment(sentence):\n",
        "\n",
        "    sentence = str(sentence).lower().split()\n",
        "    np.random.shuffle(sentence)\n",
        "    sentence = sentence[:np.random.randint(0, len(sentence))]\n",
        "\n",
        "    return \" \".join(sentence)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_OmIKxMbmWpZ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "b675d816-2215-41ab-b474-06127752639f"
      },
      "source": [
        "data.ingredients.iloc[42]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'package dry yeast water sugar salt oil flour cornmeal jar ragu pizza sauce mozzarella cheese oregano pepper pepperoni olive green pepper mushroom parmesan cheese'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eGYPjVBzmTP1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "0c44c892-59ec-4b5d-aecb-647106006f8e"
      },
      "source": [
        "augment(data.ingredients.iloc[42])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'ragu pepperoni salt mozzarella oil cheese sauce sugar pepper oregano pizza cheese dry cornmeal jar green yeast pepper mushroom olive water flour package'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GGtPn4Kwc-3O"
      },
      "source": [
        "def generator(meta_batch_siz, batch_siz, df, aug=False):\n",
        "\n",
        "    if meta_batch_siz % batch_siz != 0:\n",
        "        raise ValueError(f\"meta-batch size = {meta_batch_siz} should be divisible by batch size = {batch_siz}\")\n",
        "\n",
        "    while True:\n",
        "\n",
        "        df = df.sample(frac=1).reset_index(drop=True)\n",
        "\n",
        "        namez, ingredz, steps, labels = [], [], [], []\n",
        "\n",
        "        for index in range(len(df)):\n",
        "\n",
        "            start = 1\n",
        "\n",
        "            if aug:\n",
        "                n = np.array(tokenize_name(augment(df.name.iloc[index])), dtype=\"int32\")\n",
        "                i = np.array(tokenize_ingreds(augment(df.ingredients.iloc[index])), dtype=\"int32\")\n",
        "            else:\n",
        "                n = np.array(tokenize_name(df.name.iloc[index]), dtype=\"int32\")\n",
        "                i = np.array(tokenize_ingreds(df.ingredients.iloc[index]), dtype=\"int32\")\n",
        "\n",
        "            stepz = tokenize_steps(df.steps.iloc[index])\n",
        "\n",
        "            for idx in range(start, len([i for i in stepz if i != 0])-1):\n",
        "\n",
        "                steps.append(np.array(stepz[:idx]+[0]*(max_steps_len-idx), dtype=\"int32\"))\n",
        "                labels.append(np.array(stepz[idx], dtype=\"int32\"))\n",
        "                namez.append(n)\n",
        "                ingredz.append(i)\n",
        "\n",
        "                if len(labels) == meta_batch_siz: \n",
        "\n",
        "                    temp = list(zip(namez, ingredz, steps, labels)) \n",
        "                    np.random.shuffle(temp) \n",
        "                    namez, ingredz, steps, labels = zip(*temp) \n",
        "                    del temp\n",
        "\n",
        "                    for j in range(meta_batch_siz // batch_siz):\n",
        "                        yield ({model.inputs[0].name: np.array(namez[j*batch_siz:(j+1)*batch_siz]).reshape(batch_siz, max_name_len), \n",
        "                                model.inputs[1].name: np.array(ingredz[j*batch_siz:(j+1)*batch_siz]).reshape(batch_siz, max_ingreds_len), \n",
        "                                model.inputs[2].name: np.array(steps[j*batch_siz:(j+1)*batch_siz]).reshape(batch_siz, max_steps_len)},\n",
        "                                np.array(labels[j*batch_siz:(j+1)*batch_siz]).reshape(batch_siz, 1))\n",
        "\n",
        "                    start = idx + 1\n",
        "                    namez, ingredz, steps, labels = [], [], [], []\n",
        "\n",
        "        for j in range(0, len(labels), batch_siz):\n",
        "              a = np.array(namez[j:j+batch_siz])\n",
        "              b = np.array(ingredz[j:j+batch_siz])\n",
        "              c = np.array(steps[j:j+batch_siz])\n",
        "              d = np.array(labels[j:j+batch_siz])\n",
        "              yield ({model.inputs[0].name: a.reshape(len(a), max_name_len), \n",
        "                      model.inputs[1].name: b.reshape(len(b), max_ingreds_len), \n",
        "                      model.inputs[2].name: c.reshape(len(c), max_steps_len)},\n",
        "                      d.reshape(len(d), 1))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XCK7NYp23ZvB"
      },
      "source": [
        "import tensorflow as tf\n",
        "\n",
        "class MultiHeadSelfAttention(tf.keras.layers.Layer):\n",
        "    def __init__(self, embed_dim, num_heads=8):\n",
        "        super(MultiHeadSelfAttention, self).__init__()\n",
        "        self.embed_dim = embed_dim\n",
        "        self.num_heads = num_heads\n",
        "        if embed_dim % num_heads != 0:\n",
        "            raise ValueError(f\"embedding dimension = {embed_dim} should be divisible by number of heads = {num_heads}\")\n",
        "        self.projection_dim = embed_dim // num_heads\n",
        "        self.query_dense = tf.keras.layers.Dense(embed_dim)\n",
        "        self.key_dense = tf.keras.layers.Dense(embed_dim)\n",
        "        self.value_dense = tf.keras.layers.Dense(embed_dim)\n",
        "        self.combine_heads = tf.keras.layers.Dense(embed_dim)\n",
        "\n",
        "    def attention(self, query, key, value):\n",
        "        score = tf.matmul(query, key, transpose_b=True)\n",
        "        dim_key = tf.cast(tf.shape(key)[-1], tf.float32)\n",
        "        scaled_score = score / tf.math.sqrt(dim_key)\n",
        "        weights = tf.nn.softmax(scaled_score, axis=-1)\n",
        "        output = tf.matmul(weights, value)\n",
        "        return output, weights\n",
        "\n",
        "    def separate_heads(self, x, batch_size):\n",
        "        x = tf.reshape(x, (batch_size, -1, self.num_heads, self.projection_dim))\n",
        "        return tf.transpose(x, perm=[0, 2, 1, 3])\n",
        "\n",
        "    def call(self, inputs):\n",
        "        # x.shape = [batch_size, seq_len, embedding_dim]\n",
        "        batch_size = tf.shape(inputs)[0]\n",
        "        query = self.query_dense(inputs)  # (batch_size, seq_len, embed_dim)\n",
        "        key = self.key_dense(inputs)  # (batch_size, seq_len, embed_dim)\n",
        "        value = self.value_dense(inputs)  # (batch_size, seq_len, embed_dim)\n",
        "        query = self.separate_heads(\n",
        "            query, batch_size\n",
        "        )  # (batch_size, num_heads, seq_len, projection_dim)\n",
        "        key = self.separate_heads(\n",
        "            key, batch_size\n",
        "        )  # (batch_size, num_heads, seq_len, projection_dim)\n",
        "        value = self.separate_heads(\n",
        "            value, batch_size\n",
        "        )  # (batch_size, num_heads, seq_len, projection_dim)\n",
        "        attention, weights = self.attention(query, key, value)\n",
        "        attention = tf.transpose(\n",
        "            attention, perm=[0, 2, 1, 3]\n",
        "        )  # (batch_size, seq_len, num_heads, projection_dim)\n",
        "        concat_attention = tf.reshape(\n",
        "            attention, (batch_size, -1, self.embed_dim)\n",
        "        )  # (batch_size, seq_len, embed_dim)\n",
        "        output = self.combine_heads(\n",
        "            concat_attention\n",
        "        )  # (batch_size, seq_len, embed_dim)\n",
        "        return output\n",
        "\n",
        "    def get_config(self):\n",
        "\n",
        "        config = super().get_config().copy()\n",
        "        config.update({\n",
        "            'embed_dim': self.embed_dim,\n",
        "            'num_heads': self.num_heads,\n",
        "            'projection_dim': self.projection_dim,\n",
        "            'query_dense': self.query_dense,\n",
        "            'key_dense': self.key_dense,\n",
        "            'value_dense': self.value_dense,\n",
        "            'combine_heads': self.combine_heads,\n",
        "        })\n",
        "        return config\n",
        "\n",
        "class TransformerBlock(tf.keras.layers.Layer):\n",
        "    def __init__(self, embed_dim, num_heads, ff_dim, rate=0.1):\n",
        "        super(TransformerBlock, self).__init__()\n",
        "        self.att = MultiHeadSelfAttention(embed_dim, num_heads)\n",
        "        self.ffn = tf.keras.Sequential(\n",
        "            [tf.keras.layers.Dense(ff_dim, activation=\"relu\"), tf.keras.layers.Dense(embed_dim),]\n",
        "        )\n",
        "        self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)\n",
        "        self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)\n",
        "        self.dropout1 = tf.keras.layers.Dropout(rate)\n",
        "        self.dropout2 = tf.keras.layers.Dropout(rate)\n",
        "\n",
        "    def call(self, inputs, training):\n",
        "        attn_output = self.att(inputs)\n",
        "        attn_output = self.dropout1(attn_output, training=training)\n",
        "        out1 = self.layernorm1(inputs + attn_output)\n",
        "        ffn_output = self.ffn(out1)\n",
        "        ffn_output = self.dropout2(ffn_output, training=training)\n",
        "        return self.layernorm2(out1 + ffn_output)\n",
        "\n",
        "    def get_config(self):\n",
        "\n",
        "        config = super().get_config().copy()\n",
        "        config.update({\n",
        "            'att': self.att,\n",
        "            'ffn': self.ffn,\n",
        "            'layernorm1': self.layernorm1,\n",
        "            'layernorm2': self.layernorm2,\n",
        "            'dropout1': self.dropout1,\n",
        "            'dropout2': self.dropout2,\n",
        "        })\n",
        "        return config\n",
        "\n",
        "class TokenAndPositionEmbedding(tf.keras.layers.Layer):\n",
        "    def __init__(self, maxlen, vocab_size, embed_dim):\n",
        "        super(TokenAndPositionEmbedding, self).__init__()\n",
        "        self.token_emb = tf.keras.layers.Embedding(input_dim=vocab_size, output_dim=embed_dim)\n",
        "        self.pos_emb = tf.keras.layers.Embedding(input_dim=maxlen, output_dim=embed_dim)\n",
        "\n",
        "    def call(self, x):\n",
        "        maxlen = tf.shape(x)[-1]\n",
        "        positions = tf.range(start=0, limit=maxlen, delta=1)\n",
        "        positions = self.pos_emb(positions)\n",
        "        x = self.token_emb(x)\n",
        "        return x + positions\n",
        "\n",
        "    def get_config(self):\n",
        "\n",
        "        config = super().get_config().copy()\n",
        "        config.update({\n",
        "            'token_emb': self.token_emb,\n",
        "            'pos_emb': self.pos_emb,\n",
        "        })\n",
        "        return config"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FAs795j0Zq2t"
      },
      "source": [
        "def attention_block(prev_layer): \n",
        "\n",
        "    # prev_layer.shape = (batch_size, time_steps, input_dim)\n",
        "\n",
        "    time_steps = prev_layer.shape[1]\n",
        "    input_dim = prev_layer.shape[2]\n",
        "\n",
        "    x = tf.keras.layers.Permute((2, 1))(prev_layer) # (batch_size, input_dim, time_steps)\n",
        "    x = tf.keras.layers.Dense(time_steps, activation='softmax')(x) # attention matrix\n",
        "    x = tf.keras.layers.Permute((2, 1))(x) # (batch_size, time_steps, input_dim)\n",
        "\n",
        "    self_weighted_attention = tf.keras.layers.Multiply()([prev_layer, x])\n",
        "\n",
        "    return self_weighted_attention"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7PGRBDCnFs42",
        "outputId": "ec279a17-4fb1-4b88-f3cb-48c42b2c72a6"
      },
      "source": [
        "tf.keras.backend.clear_session()\n",
        "\n",
        "embed_dim = 128\n",
        "num_heads = 8 \n",
        "ff_dim = 512 \n",
        "\n",
        "input_name = tf.keras.layers.Input(shape=(max_name_len,), dtype=\"int32\", name='name_input')\n",
        "name_embed = TokenAndPositionEmbedding(max_name_len, elements, embed_dim)(input_name)\n",
        "name_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(name_embed)\n",
        "\n",
        "input_ingreds = tf.keras.layers.Input(shape=(max_ingreds_len,), dtype=\"int32\", name='ingreds_input')\n",
        "ingreds_embed = TokenAndPositionEmbedding(max_ingreds_len, elements, embed_dim)(input_ingreds)\n",
        "ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(ingreds_embed)\n",
        "\n",
        "concat_name_ingreds = tf.keras.layers.concatenate([name_transformer, ingreds_transformer], axis=1)\n",
        "name_ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(concat_name_ingreds)\n",
        "name_ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(name_ingreds_transformer)\n",
        "name_ingreds_pooled_output = tf.keras.layers.GlobalMaxPooling1D()(name_ingreds_transformer)\n",
        "\n",
        "input_steps = tf.keras.layers.Input(shape=(max_steps_len,), dtype=\"int32\", name='steps_input')\n",
        "steps_embed = TokenAndPositionEmbedding(max_steps_len, elements, embed_dim)(input_steps)\n",
        "steps_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(steps_embed)\n",
        "steps_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(steps_transformer)\n",
        "steps_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(steps_transformer)\n",
        "steps_pooled_output = tf.keras.layers.GlobalMaxPooling1D()(steps_transformer)\n",
        "\n",
        "concat_feats = tf.keras.layers.concatenate([name_ingreds_pooled_output, steps_pooled_output], axis=1)\n",
        "\n",
        "fc = tf.keras.layers.Dense(256, activation=tf.keras.activations.swish)(concat_feats)\n",
        "drop = tf.keras.layers.AlphaDropout(0.075)(fc)\n",
        "fc = tf.keras.layers.Dense(64, activation=tf.keras.activations.swish)(drop)\n",
        "\n",
        "output = tf.keras.layers.Dense(elements, activation=\"softmax\", name='output')(fc) \n",
        "\n",
        "model = tf.keras.models.Model(inputs=[input_name, input_ingreds, input_steps], outputs=[output])\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "name_input (InputLayer)         [(None, 24)]         0                                            \n",
            "__________________________________________________________________________________________________\n",
            "ingreds_input (InputLayer)      [(None, 64)]         0                                            \n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding (T (None, 24, 128)      3909888     name_input[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding_1  (None, 64, 128)      3915008     ingreds_input[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "steps_input (InputLayer)        [(None, 512)]        0                                            \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block (TransformerB (None, 24, 128)      198272      token_and_position_embedding[0][0\n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_1 (Transforme (None, 64, 128)      198272      token_and_position_embedding_1[0]\n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding_2  (None, 512, 128)     3972352     steps_input[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "concatenate (Concatenate)       (None, 88, 128)      0           transformer_block[0][0]          \n",
            "                                                                 transformer_block_1[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_4 (Transforme (None, 512, 128)     198272      token_and_position_embedding_2[0]\n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_2 (Transforme (None, 88, 128)      198272      concatenate[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_5 (Transforme (None, 512, 128)     198272      transformer_block_4[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_3 (Transforme (None, 88, 128)      198272      transformer_block_2[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_6 (Transforme (None, 512, 128)     198272      transformer_block_5[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "global_max_pooling1d (GlobalMax (None, 128)          0           transformer_block_3[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "global_max_pooling1d_1 (GlobalM (None, 128)          0           transformer_block_6[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "concatenate_1 (Concatenate)     (None, 256)          0           global_max_pooling1d[0][0]       \n",
            "                                                                 global_max_pooling1d_1[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_42 (Dense)                (None, 256)          65792       concatenate_1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "alpha_dropout (AlphaDropout)    (None, 256)          0           dense_42[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dense_43 (Dense)                (None, 64)           16448       alpha_dropout[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "output (Dense)                  (None, 30522)        1983930     dense_43[0][0]                   \n",
            "==================================================================================================\n",
            "Total params: 15,251,322\n",
            "Trainable params: 15,251,322\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jf_RUTuEDH2p",
        "outputId": "aae13286-1723-418b-cbae-ab0135b6925f"
      },
      "source": [
        "tf.keras.backend.clear_session()\n",
        "\n",
        "embed_dim = 128\n",
        "num_heads = 8 \n",
        "ff_dim = 512 \n",
        "\n",
        "input_name = tf.keras.layers.Input(shape=(max_name_len,), dtype=\"int32\", name='name_input')\n",
        "name_embed = TokenAndPositionEmbedding(max_name_len, elements, embed_dim)(input_name)\n",
        "name_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(name_embed)\n",
        "\n",
        "input_ingreds = tf.keras.layers.Input(shape=(max_ingreds_len,), dtype=\"int32\", name='ingreds_input')\n",
        "ingreds_embed = TokenAndPositionEmbedding(max_ingreds_len, elements, embed_dim)(input_ingreds)\n",
        "ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(ingreds_embed)\n",
        "\n",
        "concat_name_ingreds = tf.keras.layers.concatenate([name_transformer, ingreds_transformer], axis=1)\n",
        "name_ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(concat_name_ingreds)\n",
        "name_ingreds_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(name_ingreds_transformer)\n",
        "\n",
        "input_steps = tf.keras.layers.Input(shape=(max_steps_len,), dtype=\"int32\", name='steps_input')\n",
        "steps_embed = TokenAndPositionEmbedding(max_steps_len, elements, embed_dim)(input_steps)\n",
        "steps_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(steps_embed)\n",
        "steps_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(steps_transformer)\n",
        "\n",
        "concat_transformers = tf.keras.layers.concatenate([name_ingreds_transformer, steps_transformer], axis=1)\n",
        "concat_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(concat_transformers)\n",
        "concat_transformer = TransformerBlock(embed_dim, num_heads, ff_dim)(concat_transformer)\n",
        "\n",
        "pool = tf.keras.layers.GlobalAveragePooling1D()(concat_transformer)\n",
        "\n",
        "fc = tf.keras.layers.Dense(256, activation=tf.keras.activations.swish)(pool)\n",
        "drop = tf.keras.layers.AlphaDropout(0.075)(fc)\n",
        "fc = tf.keras.layers.Dense(64, activation=tf.keras.activations.swish)(drop)\n",
        "\n",
        "output = tf.keras.layers.Dense(elements, activation=\"softmax\", name='output')(fc) \n",
        "\n",
        "model = tf.keras.models.Model(inputs=[input_name, input_ingreds, input_steps], outputs=[output])\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "name_input (InputLayer)         [(None, 24)]         0                                            \n",
            "__________________________________________________________________________________________________\n",
            "ingreds_input (InputLayer)      [(None, 64)]         0                                            \n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding (T (None, 24, 128)      3909888     name_input[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding_1  (None, 64, 128)      3915008     ingreds_input[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block (TransformerB (None, 24, 128)      198272      token_and_position_embedding[0][0\n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_1 (Transforme (None, 64, 128)      198272      token_and_position_embedding_1[0]\n",
            "__________________________________________________________________________________________________\n",
            "steps_input (InputLayer)        [(None, 512)]        0                                            \n",
            "__________________________________________________________________________________________________\n",
            "concatenate (Concatenate)       (None, 88, 128)      0           transformer_block[0][0]          \n",
            "                                                                 transformer_block_1[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "token_and_position_embedding_2  (None, 512, 128)     3972352     steps_input[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_2 (Transforme (None, 88, 128)      198272      concatenate[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_4 (Transforme (None, 512, 128)     198272      token_and_position_embedding_2[0]\n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_3 (Transforme (None, 88, 128)      198272      transformer_block_2[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_5 (Transforme (None, 512, 128)     198272      transformer_block_4[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "concatenate_1 (Concatenate)     (None, 600, 128)     0           transformer_block_3[0][0]        \n",
            "                                                                 transformer_block_5[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_6 (Transforme (None, 600, 128)     198272      concatenate_1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "transformer_block_7 (Transforme (None, 600, 128)     198272      transformer_block_6[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling1d (Globa (None, 128)          0           transformer_block_7[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "dense_48 (Dense)                (None, 256)          33024       global_average_pooling1d[0][0]   \n",
            "__________________________________________________________________________________________________\n",
            "alpha_dropout (AlphaDropout)    (None, 256)          0           dense_48[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dense_49 (Dense)                (None, 64)           16448       alpha_dropout[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "output (Dense)                  (None, 30522)        1983930     dense_49[0][0]                   \n",
            "==================================================================================================\n",
            "Total params: 15,416,826\n",
            "Trainable params: 15,416,826\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j7LCLbeSX0c3"
      },
      "source": [
        "model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Si1JXRtvGF_S"
      },
      "source": [
        "meta_batch_size = 1310720\n",
        "batch_size = 128\n",
        "model.fit(generator(meta_batch_size, batch_size, data), \n",
        "          epochs=10)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "muwvrkb0Z1JO"
      },
      "source": [
        "def make_recipe(name, ingreds):\n",
        "\n",
        "    name = np.array(tokenize_name(augment(name)), dtype=\"int32\").reshape(1, max_name_len)\n",
        "    ingreds = np.array(tokenize_ingreds(augment(ingreds)), dtype=\"int32\").reshape(1, max_ingreds_len)\n",
        "\n",
        "    steps = [101]+[0]*(max_steps_len-1)\n",
        "\n",
        "    preds = [101]\n",
        "\n",
        "    for i in range(1, 512): \n",
        "\n",
        "        pred = np.argmax(model.predict({model.inputs[0].name: name, \n",
        "                         model.inputs[1].name: ingreds, \n",
        "                         model.inputs[2].name: np.array(steps, dtype=\"int32\").reshape(1, max_steps_len)}))\n",
        "        \n",
        "        if pred == 102:\n",
        "            break\n",
        "\n",
        "        preds.append(pred)\n",
        "        steps[i] = pred\n",
        "    \n",
        "    return decode(preds)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HxXZKpkjw-Pz"
      },
      "source": [
        "make_recipe(\"best chocolate chip cookies\", \"butter sugar eggs vanilla extract baking soda hot water salt flour chocolate chips walnuts\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zlGZ9atfxbgm"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}