39 실습 - 딥러닝 기반 자연어처리 - RNN 실습.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"accelerator":"GPU","colab":{"name":"39 실습 - 딥러닝 기반 자연어처리 - RNN 실습.ipynb","provenance":[{"file_id":"https://github.com/tensorflow/text/blob/master/docs/tutorials/text_classification_rnn.ipynb","timestamp":1637801002658}],"collapsed_sections":[],"toc_visible":true},"kernelspec":{"display_name":"Python 3","name":"python3"}},"cells":[{"cell_type":"markdown","metadata":{"id":"hX4n9TsbGw-f"},"source":["##### Copyright 2018 The TensorFlow Authors."]},{"cell_type":"code","metadata":{"cellView":"form","id":"0nbI5DtDGw-i","executionInfo":{"status":"ok","timestamp":1637801153518,"user_tz":-540,"elapsed":989,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"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."],"execution_count":1,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"9TnJztDZGw-n"},"source":["# Text classification with an RNN"]},{"cell_type":"markdown","metadata":{"id":"AfN3bMR5Gw-o"},"source":["<table class=\"tfo-notebook-buttons\" align=\"left\">\n","  <td>\n","    <a target=\"_blank\" href=\"https://www.tensorflow.org/text/tutorials/text_classification_rnn\"><img src=\"https://www.tensorflow.org/images/tf_logo_32px.png\" />View on TensorFlow.org</a>\n","  </td>\n","  <td>\n","    <a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/text/blob/master/docs/tutorials/text_classification_rnn.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/text/blob/master/docs/tutorials/text_classification_rnn.ipynb\"><img src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" />View source on GitHub</a>\n","  </td>\n","  <td>\n","    <a href=\"https://storage.googleapis.com/tensorflow_docs/text/docs/tutorials/text_classification_rnn.ipynb\"><img src=\"https://www.tensorflow.org/images/download_logo_32px.png\" />Download notebook</a>\n","  </td>\n","</table>"]},{"cell_type":"markdown","metadata":{"id":"lUWearf0Gw-p"},"source":["This text classification tutorial trains a [recurrent neural network](https://developers.google.com/machine-learning/glossary/#recurrent_neural_network) on the [IMDB large movie review dataset](http://ai.stanford.edu/~amaas/data/sentiment/) for sentiment analysis."]},{"cell_type":"markdown","metadata":{"id":"_2VQo4bajwUU"},"source":["## Setup"]},{"cell_type":"code","metadata":{"id":"z682XYsrjkY9","executionInfo":{"status":"ok","timestamp":1637801155894,"user_tz":-540,"elapsed":2380,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["import numpy as np\n","\n","import tensorflow_datasets as tfds\n","import tensorflow as tf\n","\n","tfds.disable_progress_bar()"],"execution_count":2,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"1rXHa-w9JZhb"},"source":["Import `matplotlib` and create a helper function to plot graphs:"]},{"cell_type":"code","metadata":{"id":"Mp1Z7P9pYRSK","executionInfo":{"status":"ok","timestamp":1637801155894,"user_tz":-540,"elapsed":5,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["import matplotlib.pyplot as plt\n","\n","\n","def plot_graphs(history, metric):\n","  plt.plot(history.history[metric])\n","  plt.plot(history.history['val_'+metric], '')\n","  plt.xlabel(\"Epochs\")\n","  plt.ylabel(metric)\n","  plt.legend([metric, 'val_'+metric])"],"execution_count":3,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"pRmMubr0jrE2"},"source":["## Setup input pipeline\n","\n","\n","The IMDB large movie review dataset is a *binary classification* dataset—all the reviews have either a *positive* or *negative* sentiment.\n","\n","Download the dataset using [TFDS](https://www.tensorflow.org/datasets). See the [loading text tutorial](https://www.tensorflow.org/tutorials/load_data/text) for details on how to load this sort of data manually.\n"]},{"cell_type":"code","metadata":{"id":"SHRwRoP2nVHX","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801223851,"user_tz":-540,"elapsed":67960,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"16dbe886-767d-4d3e-df07-0414ab130aa7"},"source":["dataset, info = tfds.load('imdb_reviews', with_info=True,\n","                          as_supervised=True)\n","train_dataset, test_dataset = dataset['train'], dataset['test']\n","\n","train_dataset.element_spec"],"execution_count":4,"outputs":[{"output_type":"stream","name":"stdout","text":["\u001b[1mDownloading and preparing dataset imdb_reviews/plain_text/1.0.0 (download: 80.23 MiB, generated: Unknown size, total: 80.23 MiB) to /root/tensorflow_datasets/imdb_reviews/plain_text/1.0.0...\u001b[0m\n","Shuffling and writing examples to /root/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteJ7ABY6/imdb_reviews-train.tfrecord\n","Shuffling and writing examples to /root/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteJ7ABY6/imdb_reviews-test.tfrecord\n","Shuffling and writing examples to /root/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteJ7ABY6/imdb_reviews-unsupervised.tfrecord\n"]},{"output_type":"stream","name":"stderr","text":["WARNING:absl:Dataset is using deprecated text encoder API which will be removed soon. Please use the plain_text version of the dataset and migrate to `tensorflow_text`.\n"]},{"output_type":"stream","name":"stdout","text":["\u001b[1mDataset imdb_reviews downloaded and prepared to /root/tensorflow_datasets/imdb_reviews/plain_text/1.0.0. Subsequent calls will reuse this data.\u001b[0m\n"]},{"output_type":"execute_result","data":{"text/plain":["(TensorSpec(shape=(), dtype=tf.string, name=None),\n"," TensorSpec(shape=(), dtype=tf.int64, name=None))"]},"metadata":{},"execution_count":4}]},{"cell_type":"markdown","metadata":{"id":"nWA4c2ir7g6p"},"source":["Initially this returns a dataset of (text, label pairs):"]},{"cell_type":"code","metadata":{"id":"vd4_BGKyurao","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801224226,"user_tz":-540,"elapsed":390,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"e26b1946-b0ff-4737-afdc-61036934527d"},"source":["for example, label in train_dataset.take(1):\n","  print('text: ', example.numpy())\n","  print('label: ', label.numpy())"],"execution_count":5,"outputs":[{"output_type":"stream","name":"stdout","text":["text:  b\"This was an absolutely terrible movie. Don't be lured in by Christopher Walken or Michael Ironside. Both are great actors, but this must simply be their worst role in history. Even their great acting could not redeem this movie's ridiculous storyline. This movie is an early nineties US propaganda piece. The most pathetic scenes were those when the Columbian rebels were making their cases for revolutions. Maria Conchita Alonso appeared phony, and her pseudo-love affair with Walken was nothing but a pathetic emotional plug in a movie that was devoid of any real meaning. I am disappointed that there are movies like this, ruining actor's like Christopher Walken's good name. I could barely sit through it.\"\n","label:  0\n"]}]},{"cell_type":"markdown","metadata":{"id":"z2qVJzcEluH_"},"source":["Next shuffle the data for training and create batches of these `(text, label)` pairs:"]},{"cell_type":"code","metadata":{"id":"dDsCaZCDYZgm","executionInfo":{"status":"ok","timestamp":1637801224226,"user_tz":-540,"elapsed":4,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["BUFFER_SIZE = 10000\n","BATCH_SIZE = 64"],"execution_count":6,"outputs":[]},{"cell_type":"code","metadata":{"id":"VznrltNOnUc5","executionInfo":{"status":"ok","timestamp":1637801224227,"user_tz":-540,"elapsed":5,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["train_dataset = train_dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE).prefetch(tf.data.AUTOTUNE)\n","test_dataset = test_dataset.batch(BATCH_SIZE).prefetch(tf.data.AUTOTUNE)"],"execution_count":7,"outputs":[]},{"cell_type":"code","metadata":{"id":"jqkvdcFv41wC","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801225195,"user_tz":-540,"elapsed":972,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"794f81ce-afff-43ea-f1dd-7f41a11886ad"},"source":["for example, label in train_dataset.take(1):\n","  print('texts: ', example.numpy()[:3])\n","  print()\n","  print('labels: ', label.numpy()[:3])"],"execution_count":8,"outputs":[{"output_type":"stream","name":"stdout","text":["texts:  [b'This utterly dull, senseless, pointless, spiritless, and dumb movie isn\\'t the final proof that the world can forget about Danny Boyle and his post-\"Trainspotting\" movies: \"The Beach\" already took care of that. What this low-budget oddity does is merely to secure his place among those who started very well but got completely lost in drugs, booze, ego, self-delusion, bad management or whatever it was that lead to this once-promising director\\'s quick demise.<br /><br />The premise is absurd: two losers (Ecclestone and some bimbo Jenna G - a rapper, likely) meet by chance and spontaneously start singing with fervour more akin to lunatic asylum inhabitants than a potential hit-making duo - which they become. A friend of theirs - an even bigger illiterate loser - becomes their manager by smashing a store window and stealing a video-camera by which he films them in \"action\", and then shows the tape to some music people who actually show interest in this garbage. Now, I know that the UK in recent years has put out incredible junk, but this is ridiculous; the music makes Oasis seem like The Beatles. During the studio recordings, the duo - Strumpet - change lyrics in every take and Ecclestone quite arrogantly tells the music biz guys to take it or leave it, and quite absurdly they do take it. Not only is the music total and utter trash, but its \"performers\" are anti-social; these NEWCOMERS are supposed to be calling the shots. It\\'s just too dumb. It\\'s plain awful.<br /><br />The dialog is unfunny and goes nowhere, and this rags-to-bitches story has no point and makes no sense. It often feels improvised - under the influence of drugs. Danny Boyle is a complete idiot. This little piece of trash is so bad it\\'s embarrassing to watch. Ecclestone\\'s I.Q. also has to be questioned for agreeing to be part of this nonsense. Whoever financed this \\xc2\\xa31000 joke should leave the movie business before they end up selling their own underwear on street corners.'\n"," b'A recent viewing of THAT\\'S ENTERTAINMENT has given me the urge to watch many of the classic MGM musicals from the forties and fifties. ANCHORS AWEIGH is certainly a lesser film than ON THE TOWN. The songs aren\\'t as good, nor is the chemistry between the characters. But the film beautifully interweaves classical favorites, such as Tchaikovsky. And the scene at the Hollywood Bowl, with Sinatra and Kelly emerging from the woods above it at the top, and then running down the steps, while dozens of pianists play on the piano, is the best scene in the film, even though the scene in which Kelly dances with Jerry Mouse is more famous. Classical music enthusiasts will no doubt identify the music the pianists are playing. Sinatra then croons, \"I Fall in Love Too Easily,\" before having his epiphany about whom he loves. The color is beautiful, Hollywood looks pretty with its mountains and pollution-free air (Can you imagine Hollywood in the twenties, let alone the mid-1940s?!), and the piano music is absolutely glorious. MGM certainly had a flair for creating lyrical moments like these.'\n"," b\"I really must watch a good movie soon, because it seems every other entry or so is something that I despise. However my history speaks, I must not tell a lie. Bobby Deerfield and everything about it sucks big green banana peels. I never thought that I would see a film thud as thunderously as this one did. Al Pacino isn't acting in this film: he's posing. There are many, many scenes of his character, who is a race car driver, just staring at the camera. He's perfectly awful. Marthe Keller is just as bad. These two are supposed to be in a love affair, and there is simply no chemistry whatsoever. Sydney Pollack directed this film? There's no trace of the genius behind Tootsie here. Is this the same man I cheered for in Eyes Wide Shut? I can hardly believe it. Save yourself a horrible movie experience. Run, don't walk, away from Bobby Deerfield.\"]\n","\n","labels:  [0 1 0]\n"]}]},{"cell_type":"markdown","metadata":{"id":"s5eWCo88voPY"},"source":["## Create the text encoder"]},{"cell_type":"markdown","metadata":{"id":"TFevcItw15P_"},"source":["The raw text loaded by `tfds` needs to be processed before it can be used in a model. The simplest way to process text for training is using the `TextVectorization` layer. This layer has many capabilities, but this tutorial sticks to the default behavior.\n","\n","Create the layer, and pass the dataset's text to the layer's `.adapt` method:"]},{"cell_type":"code","metadata":{"id":"uC25Lu1Yvuqy","executionInfo":{"status":"ok","timestamp":1637801230557,"user_tz":-540,"elapsed":5366,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["VOCAB_SIZE = 1000\n","encoder = tf.keras.layers.TextVectorization(\n","    max_tokens=VOCAB_SIZE)\n","encoder.adapt(train_dataset.map(lambda text, label: text))"],"execution_count":9,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"IuQzVBbe3Ldu"},"source":["The `.adapt` method sets the layer's vocabulary. Here are the first 20 tokens. After the padding and unknown tokens they're sorted by frequency: "]},{"cell_type":"code","metadata":{"id":"tBoyjjWg0Ac9","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801230557,"user_tz":-540,"elapsed":9,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"e41f0ad7-03e2-4f87-d2e2-16c73357250d"},"source":["vocab = np.array(encoder.get_vocabulary())\n","vocab[:20]"],"execution_count":10,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array(['', '[UNK]', 'the', 'and', 'a', 'of', 'to', 'is', 'in', 'it', 'i',\n","       'this', 'that', 'br', 'was', 'as', 'for', 'with', 'movie', 'but'],\n","      dtype='<U14')"]},"metadata":{},"execution_count":10}]},{"cell_type":"markdown","metadata":{"id":"mjId5pua3jHQ"},"source":["Once the vocabulary is set, the layer can encode text into indices. The tensors of indices are 0-padded to the longest sequence in the batch (unless you set a fixed `output_sequence_length`):"]},{"cell_type":"code","metadata":{"id":"RGc7C9WiwRWs","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801230558,"user_tz":-540,"elapsed":7,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"ef8c4c9d-835d-456e-a14e-bb1dab023c58"},"source":["encoded_example = encoder(example)[:3].numpy()\n","encoded_example"],"execution_count":11,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([[ 11,   1, 758, ...,   0,   0,   0],\n","       [  4,   1, 820, ...,   0,   0,   0],\n","       [ 10,  63, 217, ...,   0,   0,   0]])"]},"metadata":{},"execution_count":11}]},{"cell_type":"markdown","metadata":{"id":"F5cjz0bS39IN"},"source":["With the default settings, the process is not completely reversible. There are three main reasons for that:\n","\n","1. The default value for `preprocessing.TextVectorization`'s `standardize` argument is `\"lower_and_strip_punctuation\"`.\n","2. The limited vocabulary size and lack of character-based fallback results in some unknown tokens."]},{"cell_type":"code","metadata":{"id":"N_tD0QY5wXaK","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801230558,"user_tz":-540,"elapsed":5,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"3fb25fa6-c3ad-4912-efa3-7c9d91006410"},"source":["for n in range(3):\n","  print(\"Original: \", example[n].numpy())\n","  print(\"Round-trip: \", \" \".join(vocab[encoded_example[n]]))\n","  print()"],"execution_count":12,"outputs":[{"output_type":"stream","name":"stdout","text":["Original:  b'This utterly dull, senseless, pointless, spiritless, and dumb movie isn\\'t the final proof that the world can forget about Danny Boyle and his post-\"Trainspotting\" movies: \"The Beach\" already took care of that. What this low-budget oddity does is merely to secure his place among those who started very well but got completely lost in drugs, booze, ego, self-delusion, bad management or whatever it was that lead to this once-promising director\\'s quick demise.<br /><br />The premise is absurd: two losers (Ecclestone and some bimbo Jenna G - a rapper, likely) meet by chance and spontaneously start singing with fervour more akin to lunatic asylum inhabitants than a potential hit-making duo - which they become. A friend of theirs - an even bigger illiterate loser - becomes their manager by smashing a store window and stealing a video-camera by which he films them in \"action\", and then shows the tape to some music people who actually show interest in this garbage. Now, I know that the UK in recent years has put out incredible junk, but this is ridiculous; the music makes Oasis seem like The Beatles. During the studio recordings, the duo - Strumpet - change lyrics in every take and Ecclestone quite arrogantly tells the music biz guys to take it or leave it, and quite absurdly they do take it. Not only is the music total and utter trash, but its \"performers\" are anti-social; these NEWCOMERS are supposed to be calling the shots. It\\'s just too dumb. It\\'s plain awful.<br /><br />The dialog is unfunny and goes nowhere, and this rags-to-bitches story has no point and makes no sense. It often feels improvised - under the influence of drugs. Danny Boyle is a complete idiot. This little piece of trash is so bad it\\'s embarrassing to watch. Ecclestone\\'s I.Q. also has to be questioned for agreeing to be part of this nonsense. Whoever financed this \\xc2\\xa31000 joke should leave the movie business before they end up selling their own underwear on street corners.'\n","Round-trip:  this [UNK] dull [UNK] [UNK] [UNK] and dumb movie isnt the final [UNK] that the world can forget about [UNK] [UNK] and his [UNK] movies the [UNK] already took care of that what this [UNK] [UNK] does is [UNK] to [UNK] his place among those who started very well but got completely lost in [UNK] [UNK] [UNK] [UNK] bad [UNK] or whatever it was that lead to this [UNK] directors [UNK] [UNK] br the premise is [UNK] two [UNK] [UNK] and some [UNK] [UNK] [UNK] a [UNK] [UNK] meet by chance and [UNK] start [UNK] with [UNK] more [UNK] to [UNK] [UNK] [UNK] than a potential [UNK] [UNK] which they become a friend of [UNK] an even [UNK] [UNK] [UNK] becomes their [UNK] by [UNK] a [UNK] [UNK] and [UNK] a [UNK] by which he films them in action and then shows the [UNK] to some music people who actually show interest in this [UNK] now i know that the [UNK] in [UNK] years has put out [UNK] [UNK] but this is ridiculous the music makes [UNK] seem like the [UNK] during the [UNK] [UNK] the [UNK] [UNK] change [UNK] in every take and [UNK] quite [UNK] tells the music [UNK] guys to take it or leave it and quite [UNK] they do take it not only is the music total and [UNK] [UNK] but its [UNK] are [UNK] these [UNK] are supposed to be [UNK] the shots its just too dumb its [UNK] [UNK] br the dialog is [UNK] and goes [UNK] and this [UNK] story has no point and makes no sense it often feels [UNK] under the [UNK] of [UNK] [UNK] [UNK] is a complete [UNK] this little piece of [UNK] is so bad its [UNK] to watch [UNK] [UNK] also has to be [UNK] for [UNK] to be part of this [UNK] [UNK] [UNK] this [UNK] joke should leave the movie business before they end up [UNK] their own [UNK] on street [UNK]                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                              \n","\n","Original:  b'A recent viewing of THAT\\'S ENTERTAINMENT has given me the urge to watch many of the classic MGM musicals from the forties and fifties. ANCHORS AWEIGH is certainly a lesser film than ON THE TOWN. The songs aren\\'t as good, nor is the chemistry between the characters. But the film beautifully interweaves classical favorites, such as Tchaikovsky. And the scene at the Hollywood Bowl, with Sinatra and Kelly emerging from the woods above it at the top, and then running down the steps, while dozens of pianists play on the piano, is the best scene in the film, even though the scene in which Kelly dances with Jerry Mouse is more famous. Classical music enthusiasts will no doubt identify the music the pianists are playing. Sinatra then croons, \"I Fall in Love Too Easily,\" before having his epiphany about whom he loves. The color is beautiful, Hollywood looks pretty with its mountains and pollution-free air (Can you imagine Hollywood in the twenties, let alone the mid-1940s?!), and the piano music is absolutely glorious. MGM certainly had a flair for creating lyrical moments like these.'\n","Round-trip:  a [UNK] viewing of thats entertainment has given me the [UNK] to watch many of the classic [UNK] [UNK] from the [UNK] and [UNK] [UNK] [UNK] is certainly a [UNK] film than on the town the songs arent as good nor is the [UNK] between the characters but the film [UNK] [UNK] [UNK] [UNK] such as [UNK] and the scene at the hollywood [UNK] with [UNK] and [UNK] [UNK] from the [UNK] above it at the top and then running down the [UNK] while [UNK] of [UNK] play on the [UNK] is the best scene in the film even though the scene in which [UNK] [UNK] with [UNK] [UNK] is more famous [UNK] music [UNK] will no doubt [UNK] the music the [UNK] are playing [UNK] then [UNK] i fall in love too easily before having his [UNK] about whom he [UNK] the [UNK] is beautiful hollywood looks pretty with its [UNK] and [UNK] air can you imagine hollywood in the [UNK] let alone the [UNK] and the [UNK] music is absolutely [UNK] [UNK] certainly had a [UNK] for [UNK] [UNK] moments like these                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             \n","\n","Original:  b\"I really must watch a good movie soon, because it seems every other entry or so is something that I despise. However my history speaks, I must not tell a lie. Bobby Deerfield and everything about it sucks big green banana peels. I never thought that I would see a film thud as thunderously as this one did. Al Pacino isn't acting in this film: he's posing. There are many, many scenes of his character, who is a race car driver, just staring at the camera. He's perfectly awful. Marthe Keller is just as bad. These two are supposed to be in a love affair, and there is simply no chemistry whatsoever. Sydney Pollack directed this film? There's no trace of the genius behind Tootsie here. Is this the same man I cheered for in Eyes Wide Shut? I can hardly believe it. Save yourself a horrible movie experience. Run, don't walk, away from Bobby Deerfield.\"\n","Round-trip:  i really must watch a good movie soon because it seems every other [UNK] or so is something that i [UNK] however my history [UNK] i must not tell a [UNK] [UNK] [UNK] and everything about it [UNK] big [UNK] [UNK] [UNK] i never thought that i would see a film [UNK] as [UNK] as this one did [UNK] [UNK] isnt acting in this film hes [UNK] there are many many scenes of his character who is a [UNK] car [UNK] just [UNK] at the camera hes perfectly awful [UNK] [UNK] is just as bad these two are supposed to be in a love [UNK] and there is simply no [UNK] [UNK] [UNK] [UNK] directed this film theres no [UNK] of the [UNK] behind [UNK] here is this the same man i [UNK] for in eyes [UNK] [UNK] i can hardly believe it save yourself a horrible movie experience run dont [UNK] away from [UNK] [UNK]                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         \n","\n"]}]},{"cell_type":"markdown","metadata":{"id":"bjUqGVBxGw-t"},"source":["## Create the model"]},{"cell_type":"markdown","metadata":{"id":"W7zsmInBOCPO"},"source":["![A drawing of the information flow in the model](https://github.com/tensorflow/text/blob/master/docs/tutorials/images/bidirectional.png?raw=1)"]},{"cell_type":"markdown","metadata":{"id":"bgs6nnSTGw-t"},"source":["Above is a diagram of the model. \n","\n","1. This model can be build as a `tf.keras.Sequential`.\n","\n","2. The first layer is the `encoder`, which converts the text to a sequence of token indices.\n","\n","3. After the encoder is an embedding layer. An embedding layer stores one vector per word. When called, it converts the sequences of word indices to sequences of vectors. These vectors are trainable. After training (on enough data), words with similar meanings often have similar vectors.\n","\n","  This index-lookup is much more efficient than the equivalent operation of passing a one-hot encoded vector through a `tf.keras.layers.Dense` layer.\n","\n","4. A recurrent neural network (RNN) processes sequence input by iterating through the elements. RNNs pass the outputs from one timestep to their input on the next timestep.\n","\n","  The `tf.keras.layers.Bidirectional` wrapper can also be used with an RNN layer. This propagates the input forward and backwards through the RNN layer and then concatenates the final output. \n","\n","  * The main advantage of a bidirectional RNN is that the signal from the beginning of the input doesn't need to be processed all the way through every timestep to affect the output.  \n","\n","  * The main disadvantage of a bidirectional RNN is that you can't efficiently stream predictions as words are being added to the end.\n","\n","5. After the RNN has converted the sequence to a single vector the two `layers.Dense` do some final processing, and convert from this vector representation to a single logit as the classification output. \n"]},{"cell_type":"markdown","metadata":{"id":"V4fodCI7soQi"},"source":["The code to implement this is below:"]},{"cell_type":"code","metadata":{"id":"LwfoBkmRYcP3","executionInfo":{"status":"ok","timestamp":1637801233226,"user_tz":-540,"elapsed":2671,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["model = tf.keras.Sequential([\n","    encoder,\n","    tf.keras.layers.Embedding(\n","        input_dim=len(encoder.get_vocabulary()),\n","        output_dim=64,\n","        # Use masking to handle the variable sequence lengths\n","        mask_zero=True),\n","    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(64)),\n","    tf.keras.layers.Dense(64, activation='relu'),\n","    tf.keras.layers.Dense(1)\n","])"],"execution_count":13,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"QIGmIGkkouUb"},"source":["Please note that Keras sequential model is used here since all the layers in the model only have single input and produce single output. In case you want to use stateful RNN layer, you might want to build your model with Keras functional API or model subclassing so that you can retrieve and reuse the RNN layer states. Please check [Keras RNN guide](https://www.tensorflow.org/guide/keras/rnn#rnn_state_reuse) for more details."]},{"cell_type":"markdown","metadata":{"id":"kF-PsCk1LwjY"},"source":["The embedding layer [uses masking](https://www.tensorflow.org/guide/keras/masking_and_padding) to handle the varying sequence-lengths. All the layers after the `Embedding` support masking:"]},{"cell_type":"code","metadata":{"id":"87a8-CwfKebw","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801233226,"user_tz":-540,"elapsed":8,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"7b66485c-8edd-4e3e-eddb-0b81e0d4fabb"},"source":["print([layer.supports_masking for layer in model.layers])"],"execution_count":14,"outputs":[{"output_type":"stream","name":"stdout","text":["[False, True, True, True, True]\n"]}]},{"cell_type":"markdown","metadata":{"id":"ZlS0iaUIWLpI"},"source":["To confirm that this works as expected, evaluate a sentence twice. First, alone so there's no padding to mask:"]},{"cell_type":"code","metadata":{"id":"O41gw3KfWHus","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801239630,"user_tz":-540,"elapsed":6409,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"c91987a2-6cdd-45d5-eb52-30bcc92dd9c2"},"source":["# predict on a sample text without padding.\n","\n","sample_text = ('The movie was cool. The animation and the graphics '\n","               'were out of this world. I would recommend this movie.')\n","predictions = model.predict(np.array([sample_text]))\n","print(predictions[0])"],"execution_count":15,"outputs":[{"output_type":"stream","name":"stdout","text":["[0.00635519]\n"]}]},{"cell_type":"markdown","metadata":{"id":"K0VQmGnEWcuz"},"source":["Now, evaluate it again in a batch with a longer sentence. The result should be identical:"]},{"cell_type":"code","metadata":{"id":"UIgpuTeFNDzq","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637801239631,"user_tz":-540,"elapsed":15,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"acfecbdd-94ac-4e95-f34c-2f77f2b315c7"},"source":["# predict on a sample text with padding\n","\n","padding = \"the \" * 2000\n","predictions = model.predict(np.array([sample_text, padding]))\n","print(predictions[0])"],"execution_count":16,"outputs":[{"output_type":"stream","name":"stdout","text":["[0.00635519]\n"]}]},{"cell_type":"markdown","metadata":{"id":"sRI776ZcH3Tf"},"source":["Compile the Keras model to configure the training process:"]},{"cell_type":"code","metadata":{"id":"kj2xei41YZjC","executionInfo":{"status":"ok","timestamp":1637801239631,"user_tz":-540,"elapsed":12,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["model.compile(loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),\n","              optimizer=tf.keras.optimizers.Adam(1e-4),\n","              metrics=['accuracy'])"],"execution_count":17,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"zIwH3nto596k"},"source":["## Train the model"]},{"cell_type":"code","metadata":{"id":"hw86wWS4YgR2","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637802378729,"user_tz":-540,"elapsed":1139109,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"4eaa8346-3383-44f0-857e-f64d9b2f37c6"},"source":["history = model.fit(train_dataset, epochs=10,\n","                    validation_data=test_dataset,\n","                    validation_steps=30)"],"execution_count":18,"outputs":[{"output_type":"stream","name":"stdout","text":["Epoch 1/10\n","391/391 [==============================] - 108s 250ms/step - loss: 0.6438 - accuracy: 0.5778 - val_loss: 0.5955 - val_accuracy: 0.6359\n","Epoch 2/10\n","391/391 [==============================] - 94s 237ms/step - loss: 0.4554 - accuracy: 0.7894 - val_loss: 0.4054 - val_accuracy: 0.8120\n","Epoch 3/10\n","391/391 [==============================] - 92s 233ms/step - loss: 0.3733 - accuracy: 0.8381 - val_loss: 0.3577 - val_accuracy: 0.8464\n","Epoch 4/10\n","391/391 [==============================] - 93s 234ms/step - loss: 0.3394 - accuracy: 0.8562 - val_loss: 0.3401 - val_accuracy: 0.8474\n","Epoch 5/10\n","391/391 [==============================] - 93s 235ms/step - loss: 0.3201 - accuracy: 0.8646 - val_loss: 0.3283 - val_accuracy: 0.8510\n","Epoch 6/10\n","391/391 [==============================] - 92s 232ms/step - loss: 0.3132 - accuracy: 0.8662 - val_loss: 0.3280 - val_accuracy: 0.8635\n","Epoch 7/10\n","391/391 [==============================] - 93s 237ms/step - loss: 0.3052 - accuracy: 0.8694 - val_loss: 0.3313 - val_accuracy: 0.8635\n","Epoch 8/10\n","391/391 [==============================] - 93s 235ms/step - loss: 0.2998 - accuracy: 0.8716 - val_loss: 0.3307 - val_accuracy: 0.8583\n","Epoch 9/10\n","391/391 [==============================] - 92s 232ms/step - loss: 0.3008 - accuracy: 0.8714 - val_loss: 0.3221 - val_accuracy: 0.8562\n","Epoch 10/10\n","391/391 [==============================] - 91s 231ms/step - loss: 0.2974 - accuracy: 0.8734 - val_loss: 0.3183 - val_accuracy: 0.8646\n"]}]},{"cell_type":"code","metadata":{"id":"BaNbXi43YgUT","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637802427557,"user_tz":-540,"elapsed":48843,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"da5b3681-a2b5-43bc-eb1d-b43176aa39ed"},"source":["test_loss, test_acc = model.evaluate(test_dataset)\n","\n","print('Test Loss:', test_loss)\n","print('Test Accuracy:', test_acc)"],"execution_count":19,"outputs":[{"output_type":"stream","name":"stdout","text":["391/391 [==============================] - 49s 125ms/step - loss: 0.3183 - accuracy: 0.8629\n","Test Loss: 0.31833046674728394\n","Test Accuracy: 0.8629199862480164\n"]}]},{"cell_type":"code","metadata":{"id":"OZmwt_mzaQJk","colab":{"base_uri":"https://localhost:8080/","height":519},"executionInfo":{"status":"ok","timestamp":1637802428465,"user_tz":-540,"elapsed":922,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"619f3e34-36f7-45bd-8d81-21f55027ed67"},"source":["plt.figure(figsize=(16, 8))\n","plt.subplot(1, 2, 1)\n","plot_graphs(history, 'accuracy')\n","plt.ylim(None, 1)\n","plt.subplot(1, 2, 2)\n","plot_graphs(history, 'loss')\n","plt.ylim(0, None)"],"execution_count":20,"outputs":[{"output_type":"execute_result","data":{"text/plain":["(0.0, 0.6611712753772736)"]},"metadata":{},"execution_count":20},{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 1152x576 with 2 Axes>"]},"metadata":{"needs_background":"light"}}]},{"cell_type":"markdown","metadata":{"id":"DwSE_386uhxD"},"source":["Run a prediction on a new sentence:\n","\n","If the prediction is >= 0.0, it is positive else it is negative."]},{"cell_type":"code","metadata":{"id":"ZXgfQSgRW6zU","executionInfo":{"status":"ok","timestamp":1637802431681,"user_tz":-540,"elapsed":3220,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["sample_text = ('The movie was cool. The animation and the graphics '\n","               'were out of this world. I would recommend this movie.')\n","predictions = model.predict(np.array([sample_text]))"],"execution_count":21,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"7g1evcaRpTKm"},"source":["## Stack two or more LSTM layers\n","\n","Keras recurrent layers have two available modes that are controlled by the `return_sequences` constructor argument:\n","\n","* If `False` it returns only the last output for each input sequence (a 2D tensor of shape (batch_size, output_features)). This is the default, used in the previous model.\n","\n","* If `True` the full sequences of successive outputs for each timestep is returned (a 3D tensor of shape `(batch_size, timesteps, output_features)`).\n","\n","Here is what the flow of information looks like with `return_sequences=True`:\n","\n","![layered_bidirectional](https://github.com/tensorflow/text/blob/master/docs/tutorials/images/layered_bidirectional.png?raw=1)"]},{"cell_type":"markdown","metadata":{"id":"wbSClCrG1z8l"},"source":["The interesting thing about using an `RNN` with `return_sequences=True` is that the output still has 3-axes, like the input, so it can be passed to another RNN layer, like this:"]},{"cell_type":"code","metadata":{"id":"jo1jjO3vn0jo","executionInfo":{"status":"ok","timestamp":1637802436119,"user_tz":-540,"elapsed":4451,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["model = tf.keras.Sequential([\n","    encoder,\n","    tf.keras.layers.Embedding(len(encoder.get_vocabulary()), 64, mask_zero=True),\n","    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(64,  return_sequences=True)),\n","    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(32)),\n","    tf.keras.layers.Dense(64, activation='relu'),\n","    tf.keras.layers.Dropout(0.5),\n","    tf.keras.layers.Dense(1)\n","])"],"execution_count":22,"outputs":[]},{"cell_type":"code","metadata":{"id":"hEPV5jVGp-is","executionInfo":{"status":"ok","timestamp":1637802436120,"user_tz":-540,"elapsed":7,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}}},"source":["model.compile(loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),\n","              optimizer=tf.keras.optimizers.Adam(1e-4),\n","              metrics=['accuracy'])"],"execution_count":23,"outputs":[]},{"cell_type":"code","metadata":{"id":"LeSE-YjdqAeN","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637804351698,"user_tz":-540,"elapsed":1915584,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"8e948537-9bba-4dac-9fcf-8d3bd333e28f"},"source":["history = model.fit(train_dataset, epochs=10,\n","                    validation_data=test_dataset,\n","                    validation_steps=30)"],"execution_count":24,"outputs":[{"output_type":"stream","name":"stdout","text":["Epoch 1/10\n","391/391 [==============================] - 198s 455ms/step - loss: 0.6196 - accuracy: 0.5972 - val_loss: 0.4387 - val_accuracy: 0.8109\n","Epoch 2/10\n","391/391 [==============================] - 169s 430ms/step - loss: 0.3773 - accuracy: 0.8374 - val_loss: 0.3572 - val_accuracy: 0.8474\n","Epoch 3/10\n","391/391 [==============================] - 168s 428ms/step - loss: 0.3325 - accuracy: 0.8582 - val_loss: 0.3357 - val_accuracy: 0.8516\n","Epoch 4/10\n","391/391 [==============================] - 170s 431ms/step - loss: 0.3205 - accuracy: 0.8638 - val_loss: 0.3285 - val_accuracy: 0.8547\n","Epoch 5/10\n","391/391 [==============================] - 168s 428ms/step - loss: 0.3098 - accuracy: 0.8672 - val_loss: 0.3396 - val_accuracy: 0.8349\n","Epoch 6/10\n","391/391 [==============================] - 169s 431ms/step - loss: 0.3112 - accuracy: 0.8681 - val_loss: 0.3206 - val_accuracy: 0.8536\n","Epoch 7/10\n","391/391 [==============================] - 167s 425ms/step - loss: 0.3049 - accuracy: 0.8685 - val_loss: 0.3196 - val_accuracy: 0.8594\n","Epoch 8/10\n","391/391 [==============================] - 169s 429ms/step - loss: 0.3005 - accuracy: 0.8722 - val_loss: 0.3276 - val_accuracy: 0.8427\n","Epoch 9/10\n","391/391 [==============================] - 167s 425ms/step - loss: 0.3002 - accuracy: 0.8720 - val_loss: 0.3148 - val_accuracy: 0.8599\n","Epoch 10/10\n","391/391 [==============================] - 168s 427ms/step - loss: 0.2983 - accuracy: 0.8720 - val_loss: 0.3277 - val_accuracy: 0.8411\n"]}]},{"cell_type":"code","metadata":{"id":"_LdwilM1qPM3","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637804436316,"user_tz":-540,"elapsed":84630,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"add698c5-2f66-4e03-96a8-dae56fa85201"},"source":["test_loss, test_acc = model.evaluate(test_dataset)\n","\n","print('Test Loss:', test_loss)\n","print('Test Accuracy:', test_acc)"],"execution_count":25,"outputs":[{"output_type":"stream","name":"stdout","text":["391/391 [==============================] - 85s 217ms/step - loss: 0.3247 - accuracy: 0.8411\n","Test Loss: 0.3247363269329071\n","Test Accuracy: 0.8410800099372864\n"]}]},{"cell_type":"code","metadata":{"id":"ykUKnAoqbycW","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1637804442851,"user_tz":-540,"elapsed":6547,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"a4045f14-1f85-4ed0-8f49-3e92729648b3"},"source":["# predict on a sample text without padding.\n","\n","sample_text = ('The movie was not good. The animation and the graphics '\n","               'were terrible. I would not recommend this movie.')\n","predictions = model.predict(np.array([sample_text]))\n","print(predictions)"],"execution_count":26,"outputs":[{"output_type":"stream","name":"stdout","text":["[[-1.7720048]]\n"]}]},{"cell_type":"code","metadata":{"id":"_YYub0EDtwCu","colab":{"base_uri":"https://localhost:8080/","height":388},"executionInfo":{"status":"ok","timestamp":1637804444358,"user_tz":-540,"elapsed":1517,"user":{"displayName":"이민호","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiFPPatrtQJJCEfMd6D3DoTVRog9gVm7Ovj5Lex=s64","userId":"15829449822908558555"}},"outputId":"1ce5796a-635e-40e5-8a3c-98bce43518f9"},"source":["plt.figure(figsize=(16, 6))\n","plt.subplot(1, 2, 1)\n","plot_graphs(history, 'accuracy')\n","plt.subplot(1, 2, 2)\n","plot_graphs(history, 'loss')"],"execution_count":27,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 1152x432 with 2 Axes>"]},"metadata":{"needs_background":"light"}}]},{"cell_type":"markdown","metadata":{"id":"9xvpE3BaGw_V"},"source":["Check out other existing recurrent layers such as [GRU layers](https://www.tensorflow.org/api_docs/python/tf/keras/layers/GRU).\n","\n","If you're interestied in building custom RNNs, see the [Keras RNN Guide](https://www.tensorflow.org/guide/keras/rnn).\n"]}]}