l09c06_nlp_subwords.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "punL79CN7Ox6"
      },
      "source": [
        "##### Copyright 2020 The TensorFlow Authors."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "cellView": "form",
        "id": "_ckMIh7O7s6D"
      },
      "outputs": [],
      "source": [
        "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "# you may not use this file except in compliance with the License.\n",
        "# You may obtain a copy of the License at\n",
        "#\n",
        "# https://www.apache.org/licenses/LICENSE-2.0\n",
        "#\n",
        "# Unless required by applicable law or agreed to in writing, software\n",
        "# distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "# See the License for the specific language governing permissions and\n",
        "# limitations under the License."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CH5gnvxl-N3U"
      },
      "source": [
        "# What's in a (sub)word?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S5Uhzt6vVIB2"
      },
      "source": [
        "<table class=\"tfo-notebook-buttons\" align=\"left\">\n",
        "  <td>\n",
        "    <a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/examples/blob/master/courses/udacity_intro_to_tensorflow_for_deep_learning/l09c06_nlp_subwords.ipynb\"><img src=\"https://www.tensorflow.org/images/colab_logo_32px.png\" />Run in Google Colab</a>\n",
        "  </td>\n",
        "  <td>\n",
        "    <a target=\"_blank\" href=\"https://github.com/tensorflow/examples/blob/master/courses/udacity_intro_to_tensorflow_for_deep_learning/l09c06_nlp_subwords.ipynb\"><img src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" />View source on GitHub</a>\n",
        "  </td>\n",
        "</table>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ykxAKKa1Dl0s"
      },
      "source": [
        "In this colab, we'll work with subwords, or words made up of the pieces of larger words, and see how that impacts our network and related embeddings."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QQCr_NAT-g5w"
      },
      "source": [
        "## Import TensorFlow and related functions"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "Q8Wa_ZlX-mPH"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MRHk-4Te-yLJ"
      },
      "source": [
        "## Get the original dataset\n",
        "\n",
        "We'll once again use the dataset containing Amazon and Yelp reviews. This dataset was originally extracted from [here](https://www.kaggle.com/marklvl/sentiment-labelled-sentences-data-set)."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "XJAxrOLi-02C",
        "outputId": "345cd3f7-8ec3-4fc2-e83f-e6cbf7073e87",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "--2021-12-08 16:19:09--  https://drive.google.com/uc?id=13ySLC_ue6Umt9RJYSeM2t-V0kCv-4C-P\n",
            "Resolving drive.google.com (drive.google.com)... 64.233.189.100, 64.233.189.113, 64.233.189.102, ...\n",
            "Connecting to drive.google.com (drive.google.com)|64.233.189.100|:443... connected.\n",
            "HTTP request sent, awaiting response... 302 Moved Temporarily\n",
            "Location: https://doc-08-ak-docs.googleusercontent.com/docs/securesc/ha0ro937gcuc7l7deffksulhg5h7mbp1/ujjeloi32e4t4ctvv1lutktte0g2mtcq/1638980325000/11118900490791463723/*/13ySLC_ue6Umt9RJYSeM2t-V0kCv-4C-P [following]\n",
            "Warning: wildcards not supported in HTTP.\n",
            "--2021-12-08 16:19:10--  https://doc-08-ak-docs.googleusercontent.com/docs/securesc/ha0ro937gcuc7l7deffksulhg5h7mbp1/ujjeloi32e4t4ctvv1lutktte0g2mtcq/1638980325000/11118900490791463723/*/13ySLC_ue6Umt9RJYSeM2t-V0kCv-4C-P\n",
            "Resolving doc-08-ak-docs.googleusercontent.com (doc-08-ak-docs.googleusercontent.com)... 108.177.97.132, 2404:6800:4008:c00::84\n",
            "Connecting to doc-08-ak-docs.googleusercontent.com (doc-08-ak-docs.googleusercontent.com)|108.177.97.132|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 127831 (125K) [text/csv]\n",
            "Saving to: ‘/tmp/sentiment.csv’\n",
            "\n",
            "/tmp/sentiment.csv  100%[===================>] 124.83K  --.-KB/s    in 0.001s  \n",
            "\n",
            "2021-12-08 16:19:10 (96.1 MB/s) - ‘/tmp/sentiment.csv’ saved [127831/127831]\n",
            "\n"
          ]
        }
      ],
      "source": [
        "!wget --no-check-certificate \\\n",
        "    https://drive.google.com/uc?id=13ySLC_ue6Umt9RJYSeM2t-V0kCv-4C-P \\\n",
        "    -O /tmp/sentiment.csv"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "Dr-EDUKP_HBl"
      },
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "\n",
        "dataset = pd.read_csv('/tmp/sentiment.csv')\n",
        "\n",
        "# Just extract out sentences and labels first - we will create subwords here\n",
        "sentences = dataset['text'].tolist()\n",
        "labels = dataset['sentiment'].tolist()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8zut9Wng_R3B"
      },
      "source": [
        "## Create a subwords dataset\n",
        "\n",
        "We can use the existing Amazon and Yelp reviews dataset with `tensorflow_datasets`'s `SubwordTextEncoder` functionality. `SubwordTextEncoder.build_from_corpus()` will create a tokenizer for us. You could also use this functionality to get subwords from a much larger corpus of text as well, but we'll just use our existing dataset here.\n",
        "\n",
        "The Amazon and Yelp dataset we are using isn't super large, so we'll create a subword `vocab_size` of only the 1,000 most common words, as well as cutting off each subword to be at most 5 characters.\n",
        "\n",
        "Check out the related documentation [here](https://www.tensorflow.org/datasets/api_docs/python/tfds/features/text/SubwordTextEncoder#build_from_corpus)."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "id": "aElsgxia_43g"
      },
      "outputs": [],
      "source": [
        "import tensorflow_datasets as tfds\n",
        "\n",
        "vocab_size = 1000\n",
        "tokenizer = tfds.deprecated.text.SubwordTextEncoder.build_from_corpus(sentences, vocab_size, max_subword_length=5)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "id": "0XNZWGKqBDc3",
        "outputId": "479e8a4c-10ff-408e-c632-f2ca6585cdb6",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "I have to jiggle the plug to get it to line up right to get decent volume.\n",
            "[4, 31, 6, 849, 162, 450, 12, 1, 600, 438, 775, 6, 175, 14, 6, 55, 213, 159, 474, 775, 6, 175, 614, 380, 295, 148, 72, 789]\n",
            "I \n",
            "have \n",
            "to \n",
            "j\n",
            "ig\n",
            "gl\n",
            "e \n",
            "the \n",
            "pl\n",
            "ug\n",
            " \n",
            "to \n",
            "get \n",
            "it \n",
            "to \n",
            "li\n",
            "ne \n",
            "up \n",
            "right\n",
            " \n",
            "to \n",
            "get \n",
            "dec\n",
            "ent \n",
            "vo\n",
            "lu\n",
            "me\n",
            ".\n"
          ]
        }
      ],
      "source": [
        "# Check that the tokenizer works appropriately\n",
        "num = 5\n",
        "print(sentences[num])\n",
        "encoded = tokenizer.encode(sentences[num])\n",
        "print(encoded)\n",
        "# Separately print out each subword, decoded\n",
        "for i in encoded:\n",
        "  print(tokenizer.decode([i]))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gYnbqctXGKcC"
      },
      "source": [
        "## Replace sentence data with encoded subwords\n",
        "\n",
        "Now, we'll re-create the dataset to be used for training by actually encoding each of the individual sentences. This is equivalent to `text_to_sequences` with the `Tokenizer` we used in earlier exercises."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "id": "rAmql34aGfeV"
      },
      "outputs": [],
      "source": [
        "for i, sentence in enumerate(sentences):\n",
        "  sentences[i] = tokenizer.encode(sentence)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "jNnee_csG5Iz",
        "outputId": "c51f9872-2c37-4edb-c500-52518e402bcd",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[625, 677, 626, 274, 380, 633, 148, 844, 789]\n"
          ]
        }
      ],
      "source": [
        "# Check the sentences are appropriately replaced\n",
        "print(sentences[1])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zpIigjecHVkF"
      },
      "source": [
        "## Final pre-processing\n",
        "\n",
        "Before training, we still need to pad the sequences, as well as split into training and test sets."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "id": "INIFSAcEHool"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "\n",
        "max_length = 50\n",
        "trunc_type='post'\n",
        "padding_type='post'\n",
        "\n",
        "# Pad all sentences\n",
        "sentences_padded = pad_sequences(sentences, maxlen=max_length, \n",
        "                                 padding=padding_type, truncating=trunc_type)\n",
        "\n",
        "# Separate out the sentences and labels into training and test sets\n",
        "training_size = int(len(sentences) * 0.8)\n",
        "\n",
        "training_sentences = sentences_padded[0:training_size]\n",
        "testing_sentences = sentences_padded[training_size:]\n",
        "training_labels = labels[0:training_size]\n",
        "testing_labels = labels[training_size:]\n",
        "\n",
        "# Make labels into numpy arrays for use with the network later\n",
        "training_labels_final = np.array(training_labels)\n",
        "testing_labels_final = np.array(testing_labels)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QC9A-sTpPPiL"
      },
      "source": [
        "## Train a Sentiment Model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "eDKcL64IPcfy",
        "outputId": "a34776d6-8c2a-4516-98a6-6ba37dd76a14",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " embedding (Embedding)       (None, 50, 16)            16000     \n",
            "                                                                 \n",
            " global_average_pooling1d (G  (None, 16)               0         \n",
            " lobalAveragePooling1D)                                          \n",
            "                                                                 \n",
            " dense (Dense)               (None, 6)                 102       \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 1)                 7         \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 16,109\n",
            "Trainable params: 16,109\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "embedding_dim = 16\n",
        "model = tf.keras.Sequential([\n",
        "    tf.keras.layers.Embedding(vocab_size, embedding_dim, input_length=max_length),\n",
        "    tf.keras.layers.GlobalAveragePooling1D(),\n",
        "    tf.keras.layers.Dense(6, activation='relu'),\n",
        "    tf.keras.layers.Dense(1, activation='sigmoid')\n",
        "])\n",
        "\n",
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "id": "VqkMNtIeP3oz",
        "outputId": "14b943d8-d479-404a-a3e6-6521f0402959",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/30\n",
            "50/50 [==============================] - 3s 10ms/step - loss: 0.6929 - accuracy: 0.5556 - val_loss: 0.6926 - val_accuracy: 0.5238\n",
            "Epoch 2/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.6904 - accuracy: 0.6830 - val_loss: 0.6906 - val_accuracy: 0.6015\n",
            "Epoch 3/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.6867 - accuracy: 0.7627 - val_loss: 0.6885 - val_accuracy: 0.5815\n",
            "Epoch 4/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.6787 - accuracy: 0.7345 - val_loss: 0.6797 - val_accuracy: 0.6617\n",
            "Epoch 5/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.6640 - accuracy: 0.7376 - val_loss: 0.6691 - val_accuracy: 0.6667\n",
            "Epoch 6/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.6439 - accuracy: 0.7633 - val_loss: 0.6569 - val_accuracy: 0.6667\n",
            "Epoch 7/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.6201 - accuracy: 0.7884 - val_loss: 0.6408 - val_accuracy: 0.6842\n",
            "Epoch 8/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.5899 - accuracy: 0.8104 - val_loss: 0.6140 - val_accuracy: 0.7644\n",
            "Epoch 9/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.5592 - accuracy: 0.8205 - val_loss: 0.5895 - val_accuracy: 0.7820\n",
            "Epoch 10/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.5257 - accuracy: 0.8437 - val_loss: 0.5725 - val_accuracy: 0.7794\n",
            "Epoch 11/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.4931 - accuracy: 0.8449 - val_loss: 0.5519 - val_accuracy: 0.7895\n",
            "Epoch 12/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.4606 - accuracy: 0.8525 - val_loss: 0.5424 - val_accuracy: 0.7644\n",
            "Epoch 13/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.4271 - accuracy: 0.8650 - val_loss: 0.5318 - val_accuracy: 0.7544\n",
            "Epoch 14/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.3959 - accuracy: 0.8675 - val_loss: 0.5218 - val_accuracy: 0.7544\n",
            "Epoch 15/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.3694 - accuracy: 0.8782 - val_loss: 0.5206 - val_accuracy: 0.7444\n",
            "Epoch 16/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.3464 - accuracy: 0.8801 - val_loss: 0.5123 - val_accuracy: 0.7569\n",
            "Epoch 17/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.3267 - accuracy: 0.8895 - val_loss: 0.5051 - val_accuracy: 0.7519\n",
            "Epoch 18/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.3100 - accuracy: 0.8958 - val_loss: 0.4984 - val_accuracy: 0.7619\n",
            "Epoch 19/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.2928 - accuracy: 0.8989 - val_loss: 0.5037 - val_accuracy: 0.7544\n",
            "Epoch 20/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2791 - accuracy: 0.9046 - val_loss: 0.5128 - val_accuracy: 0.7519\n",
            "Epoch 21/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2660 - accuracy: 0.9065 - val_loss: 0.5085 - val_accuracy: 0.7569\n",
            "Epoch 22/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2541 - accuracy: 0.9134 - val_loss: 0.5141 - val_accuracy: 0.7519\n",
            "Epoch 23/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.2453 - accuracy: 0.9127 - val_loss: 0.5242 - val_accuracy: 0.7519\n",
            "Epoch 24/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2356 - accuracy: 0.9215 - val_loss: 0.5218 - val_accuracy: 0.7544\n",
            "Epoch 25/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2244 - accuracy: 0.9222 - val_loss: 0.5323 - val_accuracy: 0.7544\n",
            "Epoch 26/30\n",
            "50/50 [==============================] - 0s 6ms/step - loss: 0.2161 - accuracy: 0.9291 - val_loss: 0.5533 - val_accuracy: 0.7419\n",
            "Epoch 27/30\n",
            "50/50 [==============================] - 0s 6ms/step - loss: 0.2085 - accuracy: 0.9278 - val_loss: 0.5418 - val_accuracy: 0.7444\n",
            "Epoch 28/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.2010 - accuracy: 0.9322 - val_loss: 0.5577 - val_accuracy: 0.7469\n",
            "Epoch 29/30\n",
            "50/50 [==============================] - 0s 7ms/step - loss: 0.1954 - accuracy: 0.9353 - val_loss: 0.5718 - val_accuracy: 0.7469\n",
            "Epoch 30/30\n",
            "50/50 [==============================] - 0s 8ms/step - loss: 0.1863 - accuracy: 0.9397 - val_loss: 0.5712 - val_accuracy: 0.7393\n"
          ]
        }
      ],
      "source": [
        "num_epochs = 30\n",
        "model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
        "history = model.fit(training_sentences, training_labels_final, epochs=num_epochs, \n",
        "                    validation_data=(testing_sentences, testing_labels_final))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sj18M42kQkCi"
      },
      "source": [
        "## Visualize the Training Graph\n",
        "\n",
        "We can visualize the training graph below again. Does there appear to be a difference in how validation accuracy and loss is trending compared to with full words?"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {
        "id": "uy8KIMPIQlvH",
        "outputId": "d1d0695e-4411-43b1-e352-29138a2b128e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 541
        }
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "\n",
        "def plot_graphs(history, string):\n",
        "  plt.plot(history.history[string])\n",
        "  plt.plot(history.history['val_'+string])\n",
        "  plt.xlabel(\"Epochs\")\n",
        "  plt.ylabel(string)\n",
        "  plt.legend([string, 'val_'+string])\n",
        "  plt.show()\n",
        "  \n",
        "plot_graphs(history, \"accuracy\")\n",
        "plot_graphs(history, \"loss\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_m7QzouQQ1Rs"
      },
      "source": [
        "## Get files for visualizing the network\n",
        "\n",
        "Once again, you can visualize the sentiment related to all of the subwords using the below code and by heading to http://projector.tensorflow.org/ to upload and view the data.\n",
        "\n",
        "Note that the below code does have a few small changes to handle the different way text is encoded in our dataset compared to before with the built in `Tokenizer`.\n",
        "\n",
        "You may get an error like \"Number of tensors (999) do not match the number of lines in metadata (992).\" As long as you load the vectors first without error and wait a few seconds after this pops up, you will be able to click outside the file load menu and still view the visualization."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "dezs4wE5RMQu",
        "outputId": "09cbb4a8-9d61-4993-c9f4-d1c1e0fd9b8d",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(1000, 16)\n"
          ]
        }
      ],
      "source": [
        "# First get the weights of the embedding layer\n",
        "e = model.layers[0]\n",
        "weights = e.get_weights()[0]\n",
        "print(weights.shape) # shape: (vocab_size, embedding_dim)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "id": "LXKqy9Z1RSmt"
      },
      "outputs": [],
      "source": [
        "import io\n",
        "\n",
        "# Write out the embedding vectors and metadata\n",
        "out_v = io.open('vecs.tsv', 'w', encoding='utf-8')\n",
        "out_m = io.open('meta.tsv', 'w', encoding='utf-8')\n",
        "for word_num in range(0, vocab_size - 1):\n",
        "  word = tokenizer.decode([word_num])\n",
        "  embeddings = weights[word_num]\n",
        "  out_m.write(word + \"\\n\")\n",
        "  out_v.write('\\t'.join([str(x) for x in embeddings]) + \"\\n\")\n",
        "out_v.close()\n",
        "out_m.close()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "id": "v04wBMybRoGx",
        "outputId": "7c24aa3e-935a-4520-c2a4-0cc2b2c28038",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        }
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_f8f515f2-28cb-40c4-9db4-123aae887aaf\", \"vecs.tsv\", 186807)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_8e675b89-1a25-4638-8d6f-aaed62b72778\", \"meta.tsv\", 3901)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {}
        }
      ],
      "source": [
        "# Download the files\n",
        "try:\n",
        "  from google.colab import files\n",
        "except ImportError:\n",
        "  pass\n",
        "else:\n",
        "  files.download('vecs.tsv')\n",
        "  files.download('meta.tsv')"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "name": "l09c06_nlp_subwords.ipynb",
      "toc_visible": true,
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}