text-generation-with-an-rnn.ipynb

{"cells":[{"source":"<a href=\"https://www.kaggle.com/code/sharanharsoor/generate-text-like-chatgpt?scriptVersionId=136784193\" target=\"_blank\"><img align=\"left\" alt=\"Kaggle\" title=\"Open in Kaggle\" src=\"https://kaggle.com/static/images/open-in-kaggle.svg\"></a>","metadata":{},"cell_type":"markdown"},{"cell_type":"markdown","id":"1dd16eba","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","papermill":{"duration":0.022148,"end_time":"2023-07-14T16:31:24.478643","exception":false,"start_time":"2023-07-14T16:31:24.456495","status":"completed"},"tags":[]},"source":["<div class=\"alert alert-block alert-info\">\n","<h1>Introduction :  Text generation with an RNN </h1> <br/> \n","In the realm of text generation, following the advent of ChatGPT, it becomes crucial to grasp the utilization of Deep Learning architecture for generating textual content. Within this notebook, we shall delve into the tutorial on employing RNN (Recurrent Neural Network) for text generation provided by Google. The majority of the code presented herein is sourced directly from the aforementioned text generation tutorial. <br/>\n","\n","Index:\n","\n","<ol>\n","    <li><h4><a href=\"#Setup\">Setup</a></h4></li>\n","    <li><h4><a href=\"#Process\">Process the text</a></h4></li>\n","    <li><h4><a href=\"#Build\">Build The Model</a></h4></li>\n","    <li><h4><a href=\"#Model\">Try the model</a></h4></li>\n","    <li><h4><a href=\"#Train\">Train the model</a></h4></li>\n","    <li><h4><a href=\"#Generate\">Generate text</a></h4></li>    \n","    <li><h4><a href=\"#Export\">Export the generator</a></h4></li>\n","    <li><h4><a href=\"#Advance\">Advanced: Customized Training</a></h4></li>\n","</ol>\n","</div>"]},{"cell_type":"markdown","id":"527c419c","metadata":{"papermill":{"duration":0.021135,"end_time":"2023-07-14T16:31:24.521394","exception":false,"start_time":"2023-07-14T16:31:24.500259","status":"completed"},"tags":[]},"source":["### link to [Google](https://www.tensorflow.org/text/tutorials/text_generation) tutorial.\n"]},{"cell_type":"markdown","id":"db137584","metadata":{"papermill":{"duration":0.0216,"end_time":"2023-07-14T16:31:24.565048","exception":false,"start_time":"2023-07-14T16:31:24.543448","status":"completed"},"tags":[]},"source":["# 1. Setup \n","<div id=\"Setup\"></div>\n"]},{"cell_type":"markdown","id":"e1a920b6","metadata":{"papermill":{"duration":0.021169,"end_time":"2023-07-14T16:31:24.608025","exception":false,"start_time":"2023-07-14T16:31:24.586856","status":"completed"},"tags":[]},"source":["### Import TensorFlow and other libraries"]},{"cell_type":"code","execution_count":1,"id":"a19a4ddc","metadata":{"_kg_hide-output":true,"execution":{"iopub.execute_input":"2023-07-14T16:31:24.653146Z","iopub.status.busy":"2023-07-14T16:31:24.652667Z","iopub.status.idle":"2023-07-14T16:31:31.665289Z","shell.execute_reply":"2023-07-14T16:31:31.664222Z"},"papermill":{"duration":7.03846,"end_time":"2023-07-14T16:31:31.668098","exception":false,"start_time":"2023-07-14T16:31:24.629638","status":"completed"},"tags":[]},"outputs":[{"name":"stderr","output_type":"stream","text":["/opt/conda/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.16.5 and <1.23.0 is required for this version of SciPy (detected version 1.23.5\n","  warnings.warn(f\"A NumPy version >={np_minversion} and <{np_maxversion}\"\n","/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/__init__.py:98: UserWarning: unable to load libtensorflow_io_plugins.so: unable to open file: libtensorflow_io_plugins.so, from paths: ['/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/libtensorflow_io_plugins.so']\n","caused by: ['/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/libtensorflow_io_plugins.so: undefined symbol: _ZN3tsl6StatusC1EN10tensorflow5error4CodeESt17basic_string_viewIcSt11char_traitsIcEENS_14SourceLocationE']\n","  warnings.warn(f\"unable to load libtensorflow_io_plugins.so: {e}\")\n","/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/__init__.py:104: UserWarning: file system plugins are not loaded: unable to open file: libtensorflow_io.so, from paths: ['/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/libtensorflow_io.so']\n","caused by: ['/opt/conda/lib/python3.10/site-packages/tensorflow_io/python/ops/libtensorflow_io.so: undefined symbol: _ZTVN10tensorflow13GcsFileSystemE']\n","  warnings.warn(f\"file system plugins are not loaded: {e}\")\n"]}],"source":["import tensorflow as tf\n","\n","import numpy as np\n","import os\n","import time"]},{"cell_type":"markdown","id":"a7cb1406","metadata":{"execution":{"iopub.execute_input":"2023-07-13T15:17:42.959077Z","iopub.status.busy":"2023-07-13T15:17:42.958583Z","iopub.status.idle":"2023-07-13T15:17:42.966015Z","shell.execute_reply":"2023-07-13T15:17:42.964106Z","shell.execute_reply.started":"2023-07-13T15:17:42.959036Z"},"papermill":{"duration":0.021346,"end_time":"2023-07-14T16:31:31.711508","exception":false,"start_time":"2023-07-14T16:31:31.690162","status":"completed"},"tags":[]},"source":["### Download the Shakespeare dataset.\n","Change the following line to run this code on your own data.\n"]},{"cell_type":"code","execution_count":2,"id":"c38443aa","metadata":{"_kg_hide-output":true,"execution":{"iopub.execute_input":"2023-07-14T16:31:31.757614Z","iopub.status.busy":"2023-07-14T16:31:31.755884Z","iopub.status.idle":"2023-07-14T16:31:32.752796Z","shell.execute_reply":"2023-07-14T16:31:32.751701Z"},"papermill":{"duration":1.022092,"end_time":"2023-07-14T16:31:32.755344","exception":false,"start_time":"2023-07-14T16:31:31.733252","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Downloading data from https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt\n","1115394/1115394 [==============================] - 1s 1us/step\n"]}],"source":["path_to_file = tf.keras.utils.get_file('shakespeare.txt', 'https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt')"]},{"cell_type":"markdown","id":"e4bb5259","metadata":{"papermill":{"duration":0.022601,"end_time":"2023-07-14T16:31:32.801209","exception":false,"start_time":"2023-07-14T16:31:32.778608","status":"completed"},"tags":[]},"source":["### Read the data\n","First, look in the text:"]},{"cell_type":"code","execution_count":3,"id":"741a8c7d","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:32.848063Z","iopub.status.busy":"2023-07-14T16:31:32.847294Z","iopub.status.idle":"2023-07-14T16:31:32.854268Z","shell.execute_reply":"2023-07-14T16:31:32.853396Z"},"papermill":{"duration":0.032847,"end_time":"2023-07-14T16:31:32.856519","exception":false,"start_time":"2023-07-14T16:31:32.823672","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Length of text: 1115394 characters\n"]}],"source":["# Read, then decode for py2 compat.\n","text = open(path_to_file, 'rb').read().decode(encoding='utf-8')\n","# length of text is the number of characters in it\n","print(f'Length of text: {len(text)} characters')"]},{"cell_type":"code","execution_count":4,"id":"de895665","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:32.904671Z","iopub.status.busy":"2023-07-14T16:31:32.903157Z","iopub.status.idle":"2023-07-14T16:31:32.908712Z","shell.execute_reply":"2023-07-14T16:31:32.907765Z"},"papermill":{"duration":0.031388,"end_time":"2023-07-14T16:31:32.91077","exception":false,"start_time":"2023-07-14T16:31:32.879382","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["First Citizen:\n","Before we proceed any further, hear me speak.\n","\n","All:\n","Speak, speak.\n","\n","First Citizen:\n","You are all resolved rather to die than to famish?\n","\n","All:\n","Resolved. resolved.\n","\n","First Citizen:\n","First, you know Caius Marcius is chief enemy to the people.\n","\n"]}],"source":["# Take a look at the first 250 characters in text\n","print(text[:250])"]},{"cell_type":"code","execution_count":5,"id":"df8f8ab8","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:32.957826Z","iopub.status.busy":"2023-07-14T16:31:32.95738Z","iopub.status.idle":"2023-07-14T16:31:32.983892Z","shell.execute_reply":"2023-07-14T16:31:32.983043Z"},"papermill":{"duration":0.052753,"end_time":"2023-07-14T16:31:32.986078","exception":false,"start_time":"2023-07-14T16:31:32.933325","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["65 unique characters\n"]}],"source":["# The unique characters in the file\n","vocab = sorted(set(text))\n","print(f'{len(vocab)} unique characters')"]},{"cell_type":"markdown","id":"d9b74497","metadata":{"execution":{"iopub.execute_input":"2023-07-13T15:19:56.881297Z","iopub.status.busy":"2023-07-13T15:19:56.880889Z","iopub.status.idle":"2023-07-13T15:19:56.887857Z","shell.execute_reply":"2023-07-13T15:19:56.886514Z","shell.execute_reply.started":"2023-07-13T15:19:56.88125Z"},"papermill":{"duration":0.022415,"end_time":"2023-07-14T16:31:33.031062","exception":false,"start_time":"2023-07-14T16:31:33.008647","status":"completed"},"tags":[]},"source":["# 2. Process the text \n","<div id=\"Process\"></div>\n"]},{"cell_type":"markdown","id":"3d78a751","metadata":{"papermill":{"duration":0.023245,"end_time":"2023-07-14T16:31:33.077335","exception":false,"start_time":"2023-07-14T16:31:33.05409","status":"completed"},"tags":[]},"source":["Vectorize the text\n","\n","Before training, you need to convert the strings to a numerical representation. \n","\n","The `tf.keras.layers.StringLookup` layer can convert each character into a numeric ID. It just needs the text to be split into tokens first."]},{"cell_type":"code","execution_count":6,"id":"ef9f71ee","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:33.126117Z","iopub.status.busy":"2023-07-14T16:31:33.125144Z","iopub.status.idle":"2023-07-14T16:31:35.776794Z","shell.execute_reply":"2023-07-14T16:31:35.775847Z"},"papermill":{"duration":2.678154,"end_time":"2023-07-14T16:31:35.778909","exception":false,"start_time":"2023-07-14T16:31:33.100755","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["<tf.RaggedTensor [[b'a', b'b', b'c', b'd', b'e', b'f', b'g'], [b'x', b'y', b'z']]>"]},"execution_count":6,"metadata":{},"output_type":"execute_result"}],"source":["example_texts = ['abcdefg', 'xyz']\n","\n","chars = tf.strings.unicode_split(example_texts, input_encoding='UTF-8')\n","chars"]},{"cell_type":"markdown","id":"b99d39ae","metadata":{"papermill":{"duration":0.022627,"end_time":"2023-07-14T16:31:35.82492","exception":false,"start_time":"2023-07-14T16:31:35.802293","status":"completed"},"tags":[]},"source":["Now create the `tf.keras.layers.StringLookup` layer:"]},{"cell_type":"code","execution_count":7,"id":"7c0cf058","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:35.872895Z","iopub.status.busy":"2023-07-14T16:31:35.87132Z","iopub.status.idle":"2023-07-14T16:31:35.906366Z","shell.execute_reply":"2023-07-14T16:31:35.905395Z"},"papermill":{"duration":0.061058,"end_time":"2023-07-14T16:31:35.90862","exception":false,"start_time":"2023-07-14T16:31:35.847562","status":"completed"},"tags":[]},"outputs":[],"source":["ids_from_chars = tf.keras.layers.StringLookup(\n","    vocabulary=list(vocab), mask_token=None)"]},{"cell_type":"markdown","id":"aaa631d1","metadata":{"papermill":{"duration":0.02306,"end_time":"2023-07-14T16:31:35.954853","exception":false,"start_time":"2023-07-14T16:31:35.931793","status":"completed"},"tags":[]},"source":["It converts from tokens to character IDs:"]},{"cell_type":"code","execution_count":8,"id":"3a54eb00","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:36.002394Z","iopub.status.busy":"2023-07-14T16:31:36.002033Z","iopub.status.idle":"2023-07-14T16:31:36.014998Z","shell.execute_reply":"2023-07-14T16:31:36.013889Z"},"papermill":{"duration":0.039308,"end_time":"2023-07-14T16:31:36.01712","exception":false,"start_time":"2023-07-14T16:31:35.977812","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["<tf.RaggedTensor [[40, 41, 42, 43, 44, 45, 46], [63, 64, 65]]>"]},"execution_count":8,"metadata":{},"output_type":"execute_result"}],"source":["ids = ids_from_chars(chars)\n","ids"]},{"cell_type":"markdown","id":"c614f2da","metadata":{"papermill":{"duration":0.023308,"end_time":"2023-07-14T16:31:36.063636","exception":false,"start_time":"2023-07-14T16:31:36.040328","status":"completed"},"tags":[]},"source":["Since the goal of this tutorial is to generate text, it will also be important to invert this representation and recover human-readable strings from it. For this you can use `tf.keras.layers.StringLookup(..., invert=True)`.  "]},{"cell_type":"markdown","id":"307b878b","metadata":{"papermill":{"duration":0.022679,"end_time":"2023-07-14T16:31:36.10947","exception":false,"start_time":"2023-07-14T16:31:36.086791","status":"completed"},"tags":[]},"source":["Note: Here instead of passing the original vocabulary generated with `sorted(set(text))` use the `get_vocabulary()` method of the `tf.keras.layers.StringLookup` layer so that the `[UNK]` tokens is set the same way."]},{"cell_type":"code","execution_count":9,"id":"c844726d","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:36.156732Z","iopub.status.busy":"2023-07-14T16:31:36.156392Z","iopub.status.idle":"2023-07-14T16:31:36.175637Z","shell.execute_reply":"2023-07-14T16:31:36.174738Z"},"papermill":{"duration":0.045414,"end_time":"2023-07-14T16:31:36.177847","exception":false,"start_time":"2023-07-14T16:31:36.132433","status":"completed"},"tags":[]},"outputs":[],"source":["chars_from_ids = tf.keras.layers.StringLookup(\n","    vocabulary=ids_from_chars.get_vocabulary(), invert=True, mask_token=None)"]},{"cell_type":"markdown","id":"470ec999","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:00:53.263369Z","iopub.status.busy":"2023-07-13T16:00:53.262851Z","iopub.status.idle":"2023-07-13T16:00:53.270381Z","shell.execute_reply":"2023-07-13T16:00:53.269236Z","shell.execute_reply.started":"2023-07-13T16:00:53.263326Z"},"papermill":{"duration":0.022568,"end_time":"2023-07-14T16:31:36.223612","exception":false,"start_time":"2023-07-14T16:31:36.201044","status":"completed"},"tags":[]},"source":["This layer recovers the characters from the vectors of IDs, and returns them as a `tf.RaggedTensor` of characters:"]},{"cell_type":"code","execution_count":10,"id":"357ff717","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:36.273994Z","iopub.status.busy":"2023-07-14T16:31:36.273641Z","iopub.status.idle":"2023-07-14T16:31:36.282522Z","shell.execute_reply":"2023-07-14T16:31:36.281489Z"},"papermill":{"duration":0.03816,"end_time":"2023-07-14T16:31:36.284683","exception":false,"start_time":"2023-07-14T16:31:36.246523","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["<tf.RaggedTensor [[b'a', b'b', b'c', b'd', b'e', b'f', b'g'], [b'x', b'y', b'z']]>"]},"execution_count":10,"metadata":{},"output_type":"execute_result"}],"source":["chars = chars_from_ids(ids)\n","chars"]},{"cell_type":"markdown","id":"3d6d58e6","metadata":{"papermill":{"duration":0.022804,"end_time":"2023-07-14T16:31:36.330911","exception":false,"start_time":"2023-07-14T16:31:36.308107","status":"completed"},"tags":[]},"source":["You can `tf.strings.reduce_join` to join the characters back into strings. "]},{"cell_type":"code","execution_count":11,"id":"759c2ba3","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:36.378452Z","iopub.status.busy":"2023-07-14T16:31:36.378144Z","iopub.status.idle":"2023-07-14T16:31:36.382856Z","shell.execute_reply":"2023-07-14T16:31:36.381913Z"},"papermill":{"duration":0.030735,"end_time":"2023-07-14T16:31:36.38493","exception":false,"start_time":"2023-07-14T16:31:36.354195","status":"completed"},"tags":[]},"outputs":[],"source":["def text_from_ids(ids):\n","  return tf.strings.reduce_join(chars_from_ids(ids), axis=-1)"]},{"cell_type":"markdown","id":"91c392b6","metadata":{"papermill":{"duration":0.02274,"end_time":"2023-07-14T16:31:36.430708","exception":false,"start_time":"2023-07-14T16:31:36.407968","status":"completed"},"tags":[]},"source":["### The prediction task"]},{"cell_type":"markdown","id":"5e7b513c","metadata":{"papermill":{"duration":0.022711,"end_time":"2023-07-14T16:31:36.476396","exception":false,"start_time":"2023-07-14T16:31:36.453685","status":"completed"},"tags":[]},"source":["Given a character, or a sequence of characters, what is the most probable next character? This is the task you're training the model to perform. The input to the model will be a sequence of characters, and you train the model to predict the output—the following character at each time step.\n","\n","Since RNNs maintain an internal state that depends on the previously seen elements, given all the characters computed until this moment, what is the next character?\n"]},{"cell_type":"markdown","id":"145a25cc","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:01:43.414877Z","iopub.status.busy":"2023-07-13T16:01:43.414439Z","iopub.status.idle":"2023-07-13T16:01:43.425696Z","shell.execute_reply":"2023-07-13T16:01:43.42369Z","shell.execute_reply.started":"2023-07-13T16:01:43.414845Z"},"papermill":{"duration":0.022655,"end_time":"2023-07-14T16:31:36.521949","exception":false,"start_time":"2023-07-14T16:31:36.499294","status":"completed"},"tags":[]},"source":["### Create training examples and targets\n","\n","Next divide the text into example sequences. Each input sequence will contain `seq_length` characters from the text.\n","\n","For each input sequence, the corresponding targets contain the same length of text, except shifted one character to the right.\n","\n","So break the text into chunks of `seq_length+1`. For example, say `seq_length` is 4 and our text is \"Hello\". The input sequence would be \"Hell\", and the target sequence \"ello\".\n","\n","To do this first use the `tf.data.Dataset.from_tensor_slices` function to convert the text vector into a stream of character indices."]},{"cell_type":"code","execution_count":12,"id":"06502499","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:36.570991Z","iopub.status.busy":"2023-07-14T16:31:36.570634Z","iopub.status.idle":"2023-07-14T16:31:37.093922Z","shell.execute_reply":"2023-07-14T16:31:37.092748Z"},"papermill":{"duration":0.551016,"end_time":"2023-07-14T16:31:37.096436","exception":false,"start_time":"2023-07-14T16:31:36.54542","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["<tf.Tensor: shape=(1115394,), dtype=int64, numpy=array([19, 48, 57, ..., 46,  9,  1])>"]},"execution_count":12,"metadata":{},"output_type":"execute_result"}],"source":["all_ids = ids_from_chars(tf.strings.unicode_split(text, 'UTF-8'))\n","all_ids"]},{"cell_type":"code","execution_count":13,"id":"fa2eba01","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.148438Z","iopub.status.busy":"2023-07-14T16:31:37.148096Z","iopub.status.idle":"2023-07-14T16:31:37.15573Z","shell.execute_reply":"2023-07-14T16:31:37.154837Z"},"papermill":{"duration":0.03585,"end_time":"2023-07-14T16:31:37.157803","exception":false,"start_time":"2023-07-14T16:31:37.121953","status":"completed"},"tags":[]},"outputs":[],"source":["ids_dataset = tf.data.Dataset.from_tensor_slices(all_ids)"]},{"cell_type":"code","execution_count":14,"id":"afeed89a","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.206073Z","iopub.status.busy":"2023-07-14T16:31:37.205728Z","iopub.status.idle":"2023-07-14T16:31:37.268988Z","shell.execute_reply":"2023-07-14T16:31:37.267192Z"},"papermill":{"duration":0.090074,"end_time":"2023-07-14T16:31:37.271255","exception":false,"start_time":"2023-07-14T16:31:37.181181","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["F\n","i\n","r\n","s\n","t\n"," \n","C\n","i\n","t\n","i\n"]}],"source":["for ids in ids_dataset.take(10):\n","    print(chars_from_ids(ids).numpy().decode('utf-8'))"]},{"cell_type":"code","execution_count":15,"id":"6408c70c","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.319792Z","iopub.status.busy":"2023-07-14T16:31:37.319481Z","iopub.status.idle":"2023-07-14T16:31:37.323971Z","shell.execute_reply":"2023-07-14T16:31:37.322926Z"},"papermill":{"duration":0.030976,"end_time":"2023-07-14T16:31:37.326153","exception":false,"start_time":"2023-07-14T16:31:37.295177","status":"completed"},"tags":[]},"outputs":[],"source":["seq_length = 100\n"]},{"cell_type":"markdown","id":"334e6488","metadata":{"papermill":{"duration":0.023283,"end_time":"2023-07-14T16:31:37.372939","exception":false,"start_time":"2023-07-14T16:31:37.349656","status":"completed"},"tags":[]},"source":["The `batch` method lets you easily convert these individual characters to sequences of the desired size."]},{"cell_type":"code","execution_count":16,"id":"67e95aab","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.421837Z","iopub.status.busy":"2023-07-14T16:31:37.420859Z","iopub.status.idle":"2023-07-14T16:31:37.444491Z","shell.execute_reply":"2023-07-14T16:31:37.443548Z"},"papermill":{"duration":0.050067,"end_time":"2023-07-14T16:31:37.446499","exception":false,"start_time":"2023-07-14T16:31:37.396432","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["tf.Tensor(\n","[b'F' b'i' b'r' b's' b't' b' ' b'C' b'i' b't' b'i' b'z' b'e' b'n' b':'\n"," b'\\n' b'B' b'e' b'f' b'o' b'r' b'e' b' ' b'w' b'e' b' ' b'p' b'r' b'o'\n"," b'c' b'e' b'e' b'd' b' ' b'a' b'n' b'y' b' ' b'f' b'u' b'r' b't' b'h'\n"," b'e' b'r' b',' b' ' b'h' b'e' b'a' b'r' b' ' b'm' b'e' b' ' b's' b'p'\n"," b'e' b'a' b'k' b'.' b'\\n' b'\\n' b'A' b'l' b'l' b':' b'\\n' b'S' b'p' b'e'\n"," b'a' b'k' b',' b' ' b's' b'p' b'e' b'a' b'k' b'.' b'\\n' b'\\n' b'F' b'i'\n"," b'r' b's' b't' b' ' b'C' b'i' b't' b'i' b'z' b'e' b'n' b':' b'\\n' b'Y'\n"," b'o' b'u' b' '], shape=(101,), dtype=string)\n"]}],"source":["sequences = ids_dataset.batch(seq_length+1, drop_remainder=True)\n","\n","for seq in sequences.take(1):\n","  print(chars_from_ids(seq))"]},{"cell_type":"markdown","id":"ac687164","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:02:29.502225Z","iopub.status.busy":"2023-07-13T16:02:29.501756Z","iopub.status.idle":"2023-07-13T16:02:29.508979Z","shell.execute_reply":"2023-07-13T16:02:29.507653Z","shell.execute_reply.started":"2023-07-13T16:02:29.502188Z"},"papermill":{"duration":0.023427,"end_time":"2023-07-14T16:31:37.493373","exception":false,"start_time":"2023-07-14T16:31:37.469946","status":"completed"},"tags":[]},"source":["It's easier to see what this is doing if you join the tokens back into strings:"]},{"cell_type":"code","execution_count":17,"id":"e3a01ad4","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.542275Z","iopub.status.busy":"2023-07-14T16:31:37.540843Z","iopub.status.idle":"2023-07-14T16:31:37.559812Z","shell.execute_reply":"2023-07-14T16:31:37.558379Z"},"papermill":{"duration":0.045198,"end_time":"2023-07-14T16:31:37.561804","exception":false,"start_time":"2023-07-14T16:31:37.516606","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["b'First Citizen:\\nBefore we proceed any further, hear me speak.\\n\\nAll:\\nSpeak, speak.\\n\\nFirst Citizen:\\nYou '\n","b'are all resolved rather to die than to famish?\\n\\nAll:\\nResolved. resolved.\\n\\nFirst Citizen:\\nFirst, you k'\n","b\"now Caius Marcius is chief enemy to the people.\\n\\nAll:\\nWe know't, we know't.\\n\\nFirst Citizen:\\nLet us ki\"\n","b\"ll him, and we'll have corn at our own price.\\nIs't a verdict?\\n\\nAll:\\nNo more talking on't; let it be d\"\n","b'one: away, away!\\n\\nSecond Citizen:\\nOne word, good citizens.\\n\\nFirst Citizen:\\nWe are accounted poor citi'\n"]}],"source":["for seq in sequences.take(5):\n","  print(text_from_ids(seq).numpy())"]},{"cell_type":"markdown","id":"9489e2d6","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:02:46.19694Z","iopub.status.busy":"2023-07-13T16:02:46.196565Z","iopub.status.idle":"2023-07-13T16:02:46.205654Z","shell.execute_reply":"2023-07-13T16:02:46.203555Z","shell.execute_reply.started":"2023-07-13T16:02:46.196911Z"},"papermill":{"duration":0.023151,"end_time":"2023-07-14T16:31:37.60852","exception":false,"start_time":"2023-07-14T16:31:37.585369","status":"completed"},"tags":[]},"source":["For training you'll need a dataset of `(input, label)` pairs. Where `input` and \n","`label` are sequences. At each time step the input is the current character and the label is the next character. \n","\n","Here's a function that takes a sequence as input, duplicates, and shifts it to align the input and label for each timestep:"]},{"cell_type":"code","execution_count":18,"id":"d9b26fde","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.657286Z","iopub.status.busy":"2023-07-14T16:31:37.656924Z","iopub.status.idle":"2023-07-14T16:31:37.661829Z","shell.execute_reply":"2023-07-14T16:31:37.660819Z"},"papermill":{"duration":0.032169,"end_time":"2023-07-14T16:31:37.664307","exception":false,"start_time":"2023-07-14T16:31:37.632138","status":"completed"},"tags":[]},"outputs":[],"source":["def split_input_target(sequence):\n","    input_text = sequence[:-1]\n","    target_text = sequence[1:]\n","    return input_text, target_text"]},{"cell_type":"code","execution_count":19,"id":"2f9c0769","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.713034Z","iopub.status.busy":"2023-07-14T16:31:37.712699Z","iopub.status.idle":"2023-07-14T16:31:37.719233Z","shell.execute_reply":"2023-07-14T16:31:37.718344Z"},"papermill":{"duration":0.033075,"end_time":"2023-07-14T16:31:37.72126","exception":false,"start_time":"2023-07-14T16:31:37.688185","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["(['T', 'e', 'n', 's', 'o', 'r', 'f', 'l', 'o'],\n"," ['e', 'n', 's', 'o', 'r', 'f', 'l', 'o', 'w'])"]},"execution_count":19,"metadata":{},"output_type":"execute_result"}],"source":["split_input_target(list(\"Tensorflow\"))"]},{"cell_type":"code","execution_count":20,"id":"ca4d63c9","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.770371Z","iopub.status.busy":"2023-07-14T16:31:37.76963Z","iopub.status.idle":"2023-07-14T16:31:37.826821Z","shell.execute_reply":"2023-07-14T16:31:37.825913Z"},"papermill":{"duration":0.084179,"end_time":"2023-07-14T16:31:37.82909","exception":false,"start_time":"2023-07-14T16:31:37.744911","status":"completed"},"tags":[]},"outputs":[],"source":["dataset = sequences.map(split_input_target)"]},{"cell_type":"code","execution_count":21,"id":"b0cbf6ee","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:37.879028Z","iopub.status.busy":"2023-07-14T16:31:37.878026Z","iopub.status.idle":"2023-07-14T16:31:37.911071Z","shell.execute_reply":"2023-07-14T16:31:37.910049Z"},"papermill":{"duration":0.060053,"end_time":"2023-07-14T16:31:37.913314","exception":false,"start_time":"2023-07-14T16:31:37.853261","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Input : b'First Citizen:\\nBefore we proceed any further, hear me speak.\\n\\nAll:\\nSpeak, speak.\\n\\nFirst Citizen:\\nYou'\n","Target: b'irst Citizen:\\nBefore we proceed any further, hear me speak.\\n\\nAll:\\nSpeak, speak.\\n\\nFirst Citizen:\\nYou '\n"]}],"source":["for input_example, target_example in dataset.take(1):\n","    print(\"Input :\", text_from_ids(input_example).numpy())\n","    print(\"Target:\", text_from_ids(target_example).numpy())"]},{"cell_type":"markdown","id":"dda61f13","metadata":{"papermill":{"duration":0.023742,"end_time":"2023-07-14T16:31:37.961687","exception":false,"start_time":"2023-07-14T16:31:37.937945","status":"completed"},"tags":[]},"source":["### Create training batches\n","\n","You used `tf.data` to split the text into manageable sequences. But before feeding this data into the model, you need to shuffle the data and pack it into batches."]},{"cell_type":"code","execution_count":22,"id":"6ad2900a","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:38.011458Z","iopub.status.busy":"2023-07-14T16:31:38.010384Z","iopub.status.idle":"2023-07-14T16:31:38.026928Z","shell.execute_reply":"2023-07-14T16:31:38.025945Z"},"papermill":{"duration":0.043441,"end_time":"2023-07-14T16:31:38.029014","exception":false,"start_time":"2023-07-14T16:31:37.985573","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["<_PrefetchDataset element_spec=(TensorSpec(shape=(64, 100), dtype=tf.int64, name=None), TensorSpec(shape=(64, 100), dtype=tf.int64, name=None))>"]},"execution_count":22,"metadata":{},"output_type":"execute_result"}],"source":["# Batch size\n","BATCH_SIZE = 64\n","\n","# Buffer size to shuffle the dataset\n","# (TF data is designed to work with possibly infinite sequences,\n","# so it doesn't attempt to shuffle the entire sequence in memory. Instead,\n","# it maintains a buffer in which it shuffles elements).\n","BUFFER_SIZE = 10000\n","\n","dataset = (\n","    dataset\n","    .shuffle(BUFFER_SIZE)\n","    .batch(BATCH_SIZE, drop_remainder=True)\n","    .prefetch(tf.data.experimental.AUTOTUNE))\n","\n","dataset"]},{"cell_type":"markdown","id":"d4c882b5","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:03:56.61387Z","iopub.status.busy":"2023-07-13T16:03:56.61347Z","iopub.status.idle":"2023-07-13T16:03:56.622239Z","shell.execute_reply":"2023-07-13T16:03:56.619681Z","shell.execute_reply.started":"2023-07-13T16:03:56.613838Z"},"papermill":{"duration":0.023958,"end_time":"2023-07-14T16:31:38.076886","exception":false,"start_time":"2023-07-14T16:31:38.052928","status":"completed"},"tags":[]},"source":["# 3. Build The Model </h1> \n","<div id=\"Build\"></div>"]},{"cell_type":"markdown","id":"fc8716e9","metadata":{"papermill":{"duration":0.023559,"end_time":"2023-07-14T16:31:38.124247","exception":false,"start_time":"2023-07-14T16:31:38.100688","status":"completed"},"tags":[]},"source":["This section defines the model as a `keras.Model` subclass (For details see [Making new Layers and Models via subclassing](https://www.tensorflow.org/guide/keras/custom_layers_and_models)). \n","\n","This model has three layers:\n","\n","* `tf.keras.layers.Embedding`: The input layer. A trainable lookup table that will map each character-ID to a vector with `embedding_dim` dimensions;\n","* `tf.keras.layers.GRU`: A type of RNN with size `units=rnn_units` (You can also use an LSTM layer here.)\n","* `tf.keras.layers.Dense`: The output layer, with `vocab_size` outputs. It outputs one logit for each character in the vocabulary. These are the log-likelihood of each character according to the model."]},{"cell_type":"code","execution_count":23,"id":"e52f2633","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:38.174972Z","iopub.status.busy":"2023-07-14T16:31:38.173965Z","iopub.status.idle":"2023-07-14T16:31:38.180844Z","shell.execute_reply":"2023-07-14T16:31:38.179925Z"},"papermill":{"duration":0.034482,"end_time":"2023-07-14T16:31:38.182961","exception":false,"start_time":"2023-07-14T16:31:38.148479","status":"completed"},"tags":[]},"outputs":[],"source":["# Length of the vocabulary in StringLookup Layer\n","vocab_size = len(ids_from_chars.get_vocabulary())\n","\n","# The embedding dimension\n","embedding_dim = 256\n","\n","# Number of RNN units\n","rnn_units = 1024"]},{"cell_type":"code","execution_count":24,"id":"f367b632","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:38.232007Z","iopub.status.busy":"2023-07-14T16:31:38.231701Z","iopub.status.idle":"2023-07-14T16:31:38.241197Z","shell.execute_reply":"2023-07-14T16:31:38.240334Z"},"papermill":{"duration":0.036434,"end_time":"2023-07-14T16:31:38.243227","exception":false,"start_time":"2023-07-14T16:31:38.206793","status":"completed"},"tags":[]},"outputs":[],"source":["class MyModel(tf.keras.Model):\n","  def __init__(self, vocab_size, embedding_dim, rnn_units):\n","    super().__init__(self)\n","    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)\n","    self.gru = tf.keras.layers.GRU(rnn_units,\n","                                   return_sequences=True,\n","                                   return_state=True)\n","    self.dense = tf.keras.layers.Dense(vocab_size)\n","\n","  def call(self, inputs, states=None, return_state=False, training=False):\n","    x = inputs\n","    x = self.embedding(x, training=training)\n","    if states is None:\n","      states = self.gru.get_initial_state(x)\n","    x, states = self.gru(x, initial_state=states, training=training)\n","    x = self.dense(x, training=training)\n","\n","    if return_state:\n","      return x, states\n","    else:\n","      return x"]},{"cell_type":"code","execution_count":25,"id":"c94a0a6c","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:38.293931Z","iopub.status.busy":"2023-07-14T16:31:38.293627Z","iopub.status.idle":"2023-07-14T16:31:38.316933Z","shell.execute_reply":"2023-07-14T16:31:38.316018Z"},"papermill":{"duration":0.050454,"end_time":"2023-07-14T16:31:38.319101","exception":false,"start_time":"2023-07-14T16:31:38.268647","status":"completed"},"tags":[]},"outputs":[],"source":["model = MyModel(\n","    vocab_size=vocab_size,\n","    embedding_dim=embedding_dim,\n","    rnn_units=rnn_units)"]},{"cell_type":"markdown","id":"a4336cea","metadata":{"papermill":{"duration":0.024119,"end_time":"2023-07-14T16:31:38.367081","exception":false,"start_time":"2023-07-14T16:31:38.342962","status":"completed"},"tags":[]},"source":["For each character the model looks up the embedding, runs the GRU one timestep with the embedding as input, and applies the dense layer to generate logits predicting the log-likelihood of the next character:\n","\n","![A drawing of the data passing through the model](https://www.tensorflow.org/static/text/tutorials/images/text_generation_training.png)"]},{"cell_type":"markdown","id":"83166b56","metadata":{"papermill":{"duration":0.027723,"end_time":"2023-07-14T16:31:38.422874","exception":false,"start_time":"2023-07-14T16:31:38.395151","status":"completed"},"tags":[]},"source":["Note: For training you could use a `keras.Sequential` model here. To  generate text later you'll need to manage the RNN's internal state. It's simpler to include the state input and output options upfront, than it is to rearrange the model architecture later. For more details see the [Keras RNN guide](https://www.tensorflow.org/guide/keras/rnn#rnn_state_reuse)."]},{"cell_type":"markdown","id":"71fb67ec","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:06:48.671644Z","iopub.status.busy":"2023-07-13T16:06:48.67119Z","iopub.status.idle":"2023-07-13T16:06:48.67888Z","shell.execute_reply":"2023-07-13T16:06:48.677347Z","shell.execute_reply.started":"2023-07-13T16:06:48.671611Z"},"papermill":{"duration":0.027848,"end_time":"2023-07-14T16:31:38.478801","exception":false,"start_time":"2023-07-14T16:31:38.450953","status":"completed"},"tags":[]},"source":["# 4. Try the model \n","<div id=\"Model\"></div>"]},{"cell_type":"markdown","id":"d434a7a3","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:07:01.611307Z","iopub.status.busy":"2023-07-13T16:07:01.61019Z","iopub.status.idle":"2023-07-13T16:07:01.618586Z","shell.execute_reply":"2023-07-13T16:07:01.616811Z","shell.execute_reply.started":"2023-07-13T16:07:01.611251Z"},"papermill":{"duration":0.027857,"end_time":"2023-07-14T16:31:38.534729","exception":false,"start_time":"2023-07-14T16:31:38.506872","status":"completed"},"tags":[]},"source":["Now run the model to see that it behaves as expected.\n","\n","First check the shape of the output"]},{"cell_type":"code","execution_count":26,"id":"373772ef","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:38.590721Z","iopub.status.busy":"2023-07-14T16:31:38.589704Z","iopub.status.idle":"2023-07-14T16:31:43.490794Z","shell.execute_reply":"2023-07-14T16:31:43.489665Z"},"papermill":{"duration":4.930274,"end_time":"2023-07-14T16:31:43.492905","exception":false,"start_time":"2023-07-14T16:31:38.562631","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["(64, 100, 66) # (batch_size, sequence_length, vocab_size)\n"]}],"source":["for input_example_batch, target_example_batch in dataset.take(1):\n","    example_batch_predictions = model(input_example_batch)\n","    print(example_batch_predictions.shape, \"# (batch_size, sequence_length, vocab_size)\")"]},{"cell_type":"markdown","id":"604ccc81","metadata":{"papermill":{"duration":0.02367,"end_time":"2023-07-14T16:31:43.540872","exception":false,"start_time":"2023-07-14T16:31:43.517202","status":"completed"},"tags":[]},"source":["In the above example the sequence length of the input is `100` but the model can be run on inputs of any length:"]},{"cell_type":"code","execution_count":27,"id":"45eba15b","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:43.591306Z","iopub.status.busy":"2023-07-14T16:31:43.589515Z","iopub.status.idle":"2023-07-14T16:31:43.816656Z","shell.execute_reply":"2023-07-14T16:31:43.815539Z"},"papermill":{"duration":0.254812,"end_time":"2023-07-14T16:31:43.819366","exception":false,"start_time":"2023-07-14T16:31:43.564554","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Model: \"my_model\"\n","_________________________________________________________________\n"," Layer (type)                Output Shape              Param #   \n","=================================================================\n"," embedding (Embedding)       multiple                  16896     \n","                                                                 \n"," gru (GRU)                   multiple                  3938304   \n","                                                                 \n"," dense (Dense)               multiple                  67650     \n","                                                                 \n","=================================================================\n","Total params: 4,022,850\n","Trainable params: 4,022,850\n","Non-trainable params: 0\n","_________________________________________________________________\n"]}],"source":["model.summary()\n"]},{"cell_type":"markdown","id":"5970a8c2","metadata":{"papermill":{"duration":0.024771,"end_time":"2023-07-14T16:31:43.869301","exception":false,"start_time":"2023-07-14T16:31:43.84453","status":"completed"},"tags":[]},"source":["To get actual predictions from the model you need to sample from the output distribution, to get actual character indices. This distribution is defined by the logits over the character vocabulary.\n","\n","Note: It is important to _sample_ from this distribution as taking the _argmax_ of the distribution can easily get the model stuck in a loop.\n","\n","Try it for the first example in the batch:"]},{"cell_type":"code","execution_count":28,"id":"0069b1bc","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:43.920179Z","iopub.status.busy":"2023-07-14T16:31:43.919815Z","iopub.status.idle":"2023-07-14T16:31:43.931772Z","shell.execute_reply":"2023-07-14T16:31:43.930843Z"},"papermill":{"duration":0.039769,"end_time":"2023-07-14T16:31:43.933826","exception":false,"start_time":"2023-07-14T16:31:43.894057","status":"completed"},"tags":[]},"outputs":[],"source":["sampled_indices = tf.random.categorical(example_batch_predictions[0], num_samples=1)\n","sampled_indices = tf.squeeze(sampled_indices, axis=-1).numpy()"]},{"cell_type":"markdown","id":"afe55769","metadata":{"papermill":{"duration":0.025033,"end_time":"2023-07-14T16:31:43.984499","exception":false,"start_time":"2023-07-14T16:31:43.959466","status":"completed"},"tags":[]},"source":["This gives us, at each timestep, a prediction of the next character index:"]},{"cell_type":"code","execution_count":29,"id":"46419cfa","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.036577Z","iopub.status.busy":"2023-07-14T16:31:44.035606Z","iopub.status.idle":"2023-07-14T16:31:44.043156Z","shell.execute_reply":"2023-07-14T16:31:44.04214Z"},"papermill":{"duration":0.035898,"end_time":"2023-07-14T16:31:44.045247","exception":false,"start_time":"2023-07-14T16:31:44.009349","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["array([ 3, 27, 56,  5,  7, 10, 29,  5,  3, 17,  8,  6, 21, 63, 34, 61, 12,\n","       44, 22, 34, 11, 50,  5, 61, 52, 54, 43, 24, 55,  0, 29, 64, 39, 54,\n","       40, 15,  5, 16, 28,  4, 42, 46,  4, 38, 42, 47, 63, 55, 39, 25, 31,\n","       35, 30, 25, 51, 48, 57, 35, 39,  1, 25, 22, 65, 30, 56, 33, 44, 65,\n","       11, 62, 63, 39, 10, 57, 14, 49, 58, 12, 12, 50,  7, 17, 10, 20, 14,\n","       17, 33, 56, 16, 27, 38, 13,  8, 55, 37, 51, 64, 62, 57, 14])"]},"execution_count":29,"metadata":{},"output_type":"execute_result"}],"source":["sampled_indices"]},{"cell_type":"markdown","id":"90c9e034","metadata":{"papermill":{"duration":0.024839,"end_time":"2023-07-14T16:31:44.096109","exception":false,"start_time":"2023-07-14T16:31:44.07127","status":"completed"},"tags":[]},"source":["Decode these to see the text predicted by this untrained model:"]},{"cell_type":"code","execution_count":30,"id":"706e24e7","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.154895Z","iopub.status.busy":"2023-07-14T16:31:44.15228Z","iopub.status.idle":"2023-07-14T16:31:44.168134Z","shell.execute_reply":"2023-07-14T16:31:44.167011Z"},"papermill":{"duration":0.049207,"end_time":"2023-07-14T16:31:44.170227","exception":false,"start_time":"2023-07-14T16:31:44.12102","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Input:\n"," b'r hasten all the house to bed,\\nWhich heavy sorrow makes them apt unto:\\nRomeo is coming.\\n\\nNurse:\\nO Lo'\n","\n","Next Char Predictions:\n"," b\"!Nq&,3P&!D-'HxUv;eIU:k&vmodKp[UNK]PyZoaB&CO$cg$YchxpZLRVQLlirVZ\\nLIzQqTez:wxZ3rAjs;;k,D3GADTqCNY?-pXlywrA\"\n"]}],"source":["print(\"Input:\\n\", text_from_ids(input_example_batch[0]).numpy())\n","print()\n","print(\"Next Char Predictions:\\n\", text_from_ids(sampled_indices).numpy())"]},{"cell_type":"markdown","id":"98daa215","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:08:36.825983Z","iopub.status.busy":"2023-07-13T16:08:36.82552Z","iopub.status.idle":"2023-07-13T16:08:36.832942Z","shell.execute_reply":"2023-07-13T16:08:36.831491Z","shell.execute_reply.started":"2023-07-13T16:08:36.825945Z"},"papermill":{"duration":0.024942,"end_time":"2023-07-14T16:31:44.22061","exception":false,"start_time":"2023-07-14T16:31:44.195668","status":"completed"},"tags":[]},"source":["# 5. Train the model \n","<div id=\"Train\"></div>"]},{"cell_type":"markdown","id":"17238cb5","metadata":{"papermill":{"duration":0.033578,"end_time":"2023-07-14T16:31:44.279852","exception":false,"start_time":"2023-07-14T16:31:44.246274","status":"completed"},"tags":[]},"source":["At this point the problem can be treated as a standard classification problem. Given the previous RNN state, and the input this time step, predict the class of the next character."]},{"cell_type":"markdown","id":"3bd03ef3","metadata":{"papermill":{"duration":0.035045,"end_time":"2023-07-14T16:31:44.367892","exception":false,"start_time":"2023-07-14T16:31:44.332847","status":"completed"},"tags":[]},"source":["### Attach an optimizer, and a loss function\n","The standard tf.keras.losses.sparse_categorical_crossentropy loss function works in this case because it is applied across the last dimension of the predictions.\n","\n","Because your model returns logits, you need to set the from_logits flag.\n","\n","\n"]},{"cell_type":"code","execution_count":31,"id":"2bddefb4","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.420127Z","iopub.status.busy":"2023-07-14T16:31:44.419734Z","iopub.status.idle":"2023-07-14T16:31:44.424348Z","shell.execute_reply":"2023-07-14T16:31:44.423392Z"},"papermill":{"duration":0.032998,"end_time":"2023-07-14T16:31:44.42638","exception":false,"start_time":"2023-07-14T16:31:44.393382","status":"completed"},"tags":[]},"outputs":[],"source":["loss = tf.losses.SparseCategoricalCrossentropy(from_logits=True)"]},{"cell_type":"code","execution_count":32,"id":"45be5b61","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.47832Z","iopub.status.busy":"2023-07-14T16:31:44.477964Z","iopub.status.idle":"2023-07-14T16:31:44.516011Z","shell.execute_reply":"2023-07-14T16:31:44.514636Z"},"papermill":{"duration":0.066649,"end_time":"2023-07-14T16:31:44.518126","exception":false,"start_time":"2023-07-14T16:31:44.451477","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Prediction shape:  (64, 100, 66)  # (batch_size, sequence_length, vocab_size)\n","Mean loss:         tf.Tensor(4.1905828, shape=(), dtype=float32)\n"]}],"source":["example_batch_mean_loss = loss(target_example_batch, example_batch_predictions)\n","print(\"Prediction shape: \", example_batch_predictions.shape, \" # (batch_size, sequence_length, vocab_size)\")\n","print(\"Mean loss:        \", example_batch_mean_loss)"]},{"cell_type":"markdown","id":"6ea788e6","metadata":{"papermill":{"duration":0.025285,"end_time":"2023-07-14T16:31:44.569211","exception":false,"start_time":"2023-07-14T16:31:44.543926","status":"completed"},"tags":[]},"source":["A newly initialized model shouldn't be too sure of itself, the output logits should all have similar magnitudes. To confirm this you can check that the exponential of the mean loss is approximately equal to the vocabulary size. A much higher loss means the model is sure of its wrong answers, and is badly initialized:"]},{"cell_type":"code","execution_count":33,"id":"558d3cf8","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.621189Z","iopub.status.busy":"2023-07-14T16:31:44.620407Z","iopub.status.idle":"2023-07-14T16:31:44.630941Z","shell.execute_reply":"2023-07-14T16:31:44.630012Z"},"papermill":{"duration":0.03869,"end_time":"2023-07-14T16:31:44.632976","exception":false,"start_time":"2023-07-14T16:31:44.594286","status":"completed"},"tags":[]},"outputs":[{"data":{"text/plain":["66.06128"]},"execution_count":33,"metadata":{},"output_type":"execute_result"}],"source":["tf.exp(example_batch_mean_loss).numpy()"]},{"cell_type":"markdown","id":"7567801b","metadata":{"execution":{"iopub.execute_input":"2023-07-13T16:09:38.221357Z","iopub.status.busy":"2023-07-13T16:09:38.220602Z","iopub.status.idle":"2023-07-13T16:09:38.22906Z","shell.execute_reply":"2023-07-13T16:09:38.227239Z","shell.execute_reply.started":"2023-07-13T16:09:38.221319Z"},"papermill":{"duration":0.02531,"end_time":"2023-07-14T16:31:44.683557","exception":false,"start_time":"2023-07-14T16:31:44.658247","status":"completed"},"tags":[]},"source":["Configure the training procedure using the tf.keras.Model.compile method. Use tf.keras.optimizers.Adam with default arguments and the loss function."]},{"cell_type":"code","execution_count":34,"id":"b905f160","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.73545Z","iopub.status.busy":"2023-07-14T16:31:44.735112Z","iopub.status.idle":"2023-07-14T16:31:44.752252Z","shell.execute_reply":"2023-07-14T16:31:44.751355Z"},"papermill":{"duration":0.045554,"end_time":"2023-07-14T16:31:44.754267","exception":false,"start_time":"2023-07-14T16:31:44.708713","status":"completed"},"tags":[]},"outputs":[],"source":["model.compile(optimizer='adam', loss=loss)"]},{"cell_type":"markdown","id":"1f425018","metadata":{"papermill":{"duration":0.024961,"end_time":"2023-07-14T16:31:44.804578","exception":false,"start_time":"2023-07-14T16:31:44.779617","status":"completed"},"tags":[]},"source":["### Configure checkpoints"]},{"cell_type":"markdown","id":"24585894","metadata":{"papermill":{"duration":0.02507,"end_time":"2023-07-14T16:31:44.855194","exception":false,"start_time":"2023-07-14T16:31:44.830124","status":"completed"},"tags":[]},"source":["Use a `tf.keras.callbacks.ModelCheckpoint` to ensure that checkpoints are saved during training:"]},{"cell_type":"code","execution_count":35,"id":"de96f4d0","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:44.908311Z","iopub.status.busy":"2023-07-14T16:31:44.907372Z","iopub.status.idle":"2023-07-14T16:31:44.91316Z","shell.execute_reply":"2023-07-14T16:31:44.912266Z"},"papermill":{"duration":0.034786,"end_time":"2023-07-14T16:31:44.915176","exception":false,"start_time":"2023-07-14T16:31:44.88039","status":"completed"},"tags":[]},"outputs":[],"source":["# Directory where the checkpoints will be saved\n","checkpoint_dir = './training_checkpoints'\n","# Name of the checkpoint files\n","checkpoint_prefix = os.path.join(checkpoint_dir, \"ckpt_{epoch}\")\n","\n","checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(\n","    filepath=checkpoint_prefix,\n","    save_weights_only=True)"]},{"cell_type":"markdown","id":"60485d99","metadata":{"papermill":{"duration":0.025321,"end_time":"2023-07-14T16:31:44.965838","exception":false,"start_time":"2023-07-14T16:31:44.940517","status":"completed"},"tags":[]},"source":["### Execute the training\n","To keep training time reasonable, use 10 epochs to train the model. In Colab, set the runtime to GPU for faster training."]},{"cell_type":"code","execution_count":36,"id":"4ba1abc3","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:45.019469Z","iopub.status.busy":"2023-07-14T16:31:45.01848Z","iopub.status.idle":"2023-07-14T16:31:45.023336Z","shell.execute_reply":"2023-07-14T16:31:45.022379Z"},"papermill":{"duration":0.033553,"end_time":"2023-07-14T16:31:45.025409","exception":false,"start_time":"2023-07-14T16:31:44.991856","status":"completed"},"tags":[]},"outputs":[],"source":["EPOCHS = 20"]},{"cell_type":"code","execution_count":37,"id":"97564175","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:31:45.077915Z","iopub.status.busy":"2023-07-14T16:31:45.077112Z","iopub.status.idle":"2023-07-14T16:34:50.383014Z","shell.execute_reply":"2023-07-14T16:34:50.382047Z"},"papermill":{"duration":185.334324,"end_time":"2023-07-14T16:34:50.385178","exception":false,"start_time":"2023-07-14T16:31:45.050854","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1/20\n","172/172 [==============================] - 13s 44ms/step - loss: 2.7149\n","Epoch 2/20\n","172/172 [==============================] - 9s 39ms/step - loss: 1.9899\n","Epoch 3/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.7223\n","Epoch 4/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.5602\n","Epoch 5/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.4580\n","Epoch 6/20\n","172/172 [==============================] - 9s 40ms/step - loss: 1.3885\n","Epoch 7/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.3356\n","Epoch 8/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.2903\n","Epoch 9/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.2490\n","Epoch 10/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.2107\n","Epoch 11/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.1725\n","Epoch 12/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.1332\n","Epoch 13/20\n","172/172 [==============================] - 9s 39ms/step - loss: 1.0910\n","Epoch 14/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.0480\n","Epoch 15/20\n","172/172 [==============================] - 9s 38ms/step - loss: 1.0018\n","Epoch 16/20\n","172/172 [==============================] - 9s 39ms/step - loss: 0.9524\n","Epoch 17/20\n","172/172 [==============================] - 9s 38ms/step - loss: 0.9019\n","Epoch 18/20\n","172/172 [==============================] - 9s 38ms/step - loss: 0.8491\n","Epoch 19/20\n","172/172 [==============================] - 9s 38ms/step - loss: 0.7969\n","Epoch 20/20\n","172/172 [==============================] - 9s 39ms/step - loss: 0.7473\n"]}],"source":["history = model.fit(dataset, epochs=EPOCHS, callbacks=[checkpoint_callback])"]},{"cell_type":"markdown","id":"3d589114","metadata":{"papermill":{"duration":0.158143,"end_time":"2023-07-14T16:34:50.703721","exception":false,"start_time":"2023-07-14T16:34:50.545578","status":"completed"},"tags":[]},"source":["# 6. Generate text \n","<div id=\"Generate\"></div>"]},{"cell_type":"markdown","id":"5594e429","metadata":{"papermill":{"duration":0.157944,"end_time":"2023-07-14T16:34:51.02196","exception":false,"start_time":"2023-07-14T16:34:50.864016","status":"completed"},"tags":[]},"source":["The simplest way to generate text with this model is to run it in a loop, and keep track of the model's internal state as you execute it.\n","\n","![To generate text the model's output is fed back to the input](https://www.tensorflow.org/static/text/tutorials/images/text_generation_sampling.png)\n","\n","Each time you call the model you pass in some text and an internal state. The model returns a prediction for the next character and its new state. Pass the prediction and state back in to continue generating text.\n","\n","Each time you call the model you pass in some text and an internal state. The model returns a prediction for the next character and its new state. Pass the prediction and state back in to continue generating text.\n"]},{"cell_type":"markdown","id":"1e90b9db","metadata":{"papermill":{"duration":0.163829,"end_time":"2023-07-14T16:34:51.34283","exception":false,"start_time":"2023-07-14T16:34:51.179001","status":"completed"},"tags":[]},"source":["The following makes a single step prediction:"]},{"cell_type":"code","execution_count":38,"id":"a391fe0a","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:34:51.662061Z","iopub.status.busy":"2023-07-14T16:34:51.661686Z","iopub.status.idle":"2023-07-14T16:34:51.672471Z","shell.execute_reply":"2023-07-14T16:34:51.671802Z"},"papermill":{"duration":0.172386,"end_time":"2023-07-14T16:34:51.674287","exception":false,"start_time":"2023-07-14T16:34:51.501901","status":"completed"},"tags":[]},"outputs":[],"source":["class OneStep(tf.keras.Model):\n","  def __init__(self, model, chars_from_ids, ids_from_chars, temperature=1.0):\n","    super().__init__()\n","    self.temperature = temperature\n","    self.model = model\n","    self.chars_from_ids = chars_from_ids\n","    self.ids_from_chars = ids_from_chars\n","\n","    # Create a mask to prevent \"[UNK]\" from being generated.\n","    skip_ids = self.ids_from_chars(['[UNK]'])[:, None]\n","    sparse_mask = tf.SparseTensor(\n","        # Put a -inf at each bad index.\n","        values=[-float('inf')]*len(skip_ids),\n","        indices=skip_ids,\n","        # Match the shape to the vocabulary\n","        dense_shape=[len(ids_from_chars.get_vocabulary())])\n","    self.prediction_mask = tf.sparse.to_dense(sparse_mask)\n","\n","  @tf.function\n","  def generate_one_step(self, inputs, states=None):\n","    # Convert strings to token IDs.\n","    input_chars = tf.strings.unicode_split(inputs, 'UTF-8')\n","    input_ids = self.ids_from_chars(input_chars).to_tensor()\n","\n","    # Run the model.\n","    # predicted_logits.shape is [batch, char, next_char_logits]\n","    predicted_logits, states = self.model(inputs=input_ids, states=states,\n","                                          return_state=True)\n","    # Only use the last prediction.\n","    predicted_logits = predicted_logits[:, -1, :]\n","    predicted_logits = predicted_logits/self.temperature\n","    # Apply the prediction mask: prevent \"[UNK]\" from being generated.\n","    predicted_logits = predicted_logits + self.prediction_mask\n","\n","    # Sample the output logits to generate token IDs.\n","    predicted_ids = tf.random.categorical(predicted_logits, num_samples=1)\n","    predicted_ids = tf.squeeze(predicted_ids, axis=-1)\n","\n","    # Convert from token ids to characters\n","    predicted_chars = self.chars_from_ids(predicted_ids)\n","\n","    # Return the characters and model state.\n","    return predicted_chars, states"]},{"cell_type":"code","execution_count":39,"id":"5a92e9ae","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:34:51.99495Z","iopub.status.busy":"2023-07-14T16:34:51.993935Z","iopub.status.idle":"2023-07-14T16:34:52.015156Z","shell.execute_reply":"2023-07-14T16:34:52.014221Z"},"papermill":{"duration":0.183768,"end_time":"2023-07-14T16:34:52.017251","exception":false,"start_time":"2023-07-14T16:34:51.833483","status":"completed"},"tags":[]},"outputs":[],"source":["one_step_model = OneStep(model, chars_from_ids, ids_from_chars)"]},{"cell_type":"markdown","id":"a0ba7462","metadata":{"papermill":{"duration":0.160553,"end_time":"2023-07-14T16:34:52.336061","exception":false,"start_time":"2023-07-14T16:34:52.175508","status":"completed"},"tags":[]},"source":["Run it in a loop to generate some text. Looking at the generated text, you'll see the model knows when to capitalize, make paragraphs and imitates a Shakespeare-like writing vocabulary. With the small number of training epochs, it has not yet learned to form coherent sentences."]},{"cell_type":"code","execution_count":40,"id":"7fb57953","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:34:52.657751Z","iopub.status.busy":"2023-07-14T16:34:52.656745Z","iopub.status.idle":"2023-07-14T16:34:55.885164Z","shell.execute_reply":"2023-07-14T16:34:55.884193Z"},"papermill":{"duration":3.392156,"end_time":"2023-07-14T16:34:55.88726","exception":false,"start_time":"2023-07-14T16:34:52.495104","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["ROMEO:\n","\n","MENENIUS:\n","All lost is gently, as you call.\n","\n","GLOUCESTER:\n","The commodity of such comes! Pray Go tent.\n","My soul sich a sweet world do not see thou couct\n","When bloody treadunate, till a crown.\n","Twice we potter like men that were most also,e.\n","Peace, but not becomes him not again.\n","Withal, away! be putting me to make your majesty\n","Give way the charge had left us rubuted within, my lord\n","Shame to your androses: I would be keen,\n","Though 'twere to become my onnation; he hath been\n","as a grave,--\n","\n","CORIOLANUS:\n","Your peace is face.\n","\n","PROSPERO:\n","Hence use thee, man; the thorns we seems it;\n","He trembling gall against the fiery of your good\n","frier than o'er a bawd. Pils me present\n","These duty as his. 'Houth fiery hours! A toot!\n","O man, ho! no, good my lord, and near I lose you\n","As you do wrong.\n","\n","ROMEO:\n","Thou'rt by no man's, I had rather look upon\n","And leave their unfeigned faults, and trust I fear:\n","And, if you would say, sir, but a wrink i' the priest,\n","Pinched all their headstrong prince: our sestimance vew\n","I told the \n","\n","________________________________________________________________________________\n","\n","Run time: 3.2167346477508545\n"]}],"source":["start = time.time()\n","states = None\n","next_char = tf.constant(['ROMEO:'])\n","result = [next_char]\n","\n","for n in range(1000):\n","  next_char, states = one_step_model.generate_one_step(next_char, states=states)\n","  result.append(next_char)\n","\n","result = tf.strings.join(result)\n","end = time.time()\n","print(result[0].numpy().decode('utf-8'), '\\n\\n' + '_'*80)\n","print('\\nRun time:', end - start)"]},{"cell_type":"markdown","id":"2ea38165","metadata":{"papermill":{"duration":0.159445,"end_time":"2023-07-14T16:34:56.206336","exception":false,"start_time":"2023-07-14T16:34:56.046891","status":"completed"},"tags":[]},"source":["The easiest thing you can do to improve the results is to train it for longer (try EPOCHS = 30).\n","\n","You can also experiment with a different start string, try adding another RNN layer to improve the model's accuracy, or adjust the temperature parameter to generate more or less random predictions."]},{"cell_type":"markdown","id":"2a823d71","metadata":{"papermill":{"duration":0.1597,"end_time":"2023-07-14T16:34:56.527139","exception":false,"start_time":"2023-07-14T16:34:56.367439","status":"completed"},"tags":[]},"source":["If you want the model to generate text faster the easiest thing you can do is batch the text generation. In the example below the model generates 5 outputs in about the same time it took to generate 1 above."]},{"cell_type":"code","execution_count":41,"id":"62cb4eda","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:34:56.853224Z","iopub.status.busy":"2023-07-14T16:34:56.852847Z","iopub.status.idle":"2023-07-14T16:34:59.794252Z","shell.execute_reply":"2023-07-14T16:34:59.793202Z"},"papermill":{"duration":3.103875,"end_time":"2023-07-14T16:34:59.796286","exception":false,"start_time":"2023-07-14T16:34:56.692411","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["tf.Tensor(\n","[b\"ROMEO:\\nThe day, since we have very thanks.\\n\\nMAMILLIUS:\\nIndeed, look thee? whether nothing came it: brother,\\nAs blambs so, even in men stood for the\\nwhich time speak into a wonder famous are.\\n\\nGRUMIO:\\nNay, your nor fare would payn the foe.\\n\\nGador:\\nWe were not for what pearls be slipp'd off the king's shewh,\\nAnd pace the stroke before the weal; once more,\\nIs he enjoin'd to any Baptist.\\n\\nISABELLA:\\nWhat, you may, sir: I know this weal o' the feast.\\nO woful, widow! a durbers? down?\\n\\nMONTAGUE:\\nAnd I am sold it govern, you might peace,\\nThat it may call the tempet of this death,\\nHath lost apasely; but then I'll fight for ready.\\nWill you perceive me not! another hundred them,\\nBut by-shaped things take away the drum,\\nWhich he shall blush for treed, prepared grim her business\\nCagal gold in is: and that will speak for me.\\n\\nDUKE VINCENTIO:\\nHe stalls, he'll not be tongue!\\n\\nGLOUCESTER:\\nLet me discover, my gracious lay;\\nShe'll be the jest, on whom a deatheath profess,\\nHath nothing blue instructions to disco\"\n"," b\"ROMEO:\\nVery holds above me to be a burth'd was again.\\nWelcome home!\\n\\nMENENIUS:\\nIt leave of shame; for when I parted with monarchy,\\nAnd, from his friends plots by provides fearch; by this again,\\nAnothed but beginning country's to den;\\nAnd leave you only pains; but I do wrath in sure.\\nHark, our dear lord of God's name: hast thou facch,\\nWill bring along Tuts of folfuary\\nA great man's work, take up their success,\\nOf which thy yits doth long rather vew the queen,\\nShe that isle keepers Catesby hath about me\\nTo reap the hear, all sears upon thee.\\nThe new-denerate raise alick, that will seem to out,\\nAs well as pearls as exercise. Look, like, you know\\nO'er spectable days may not do't.\\n\\nBENVOLIO:\\nWell, then, see we? the duke hath not encounter me\\nBecause I will from thee hence.\\n\\nFirst Marship:\\n\\nBUCKINGHAM:\\nI will not grink too for money of the house:\\nHe Addes distunguis. Here cousin York, the fring is dead;\\nWhich in another's glories plague executioner,\\nConsorted times it in the seat, this music, as h\"\n"," b\"ROMEO:\\nMy seab is well, my lords. Go, play, good Petruchio!\\nBut, O, this slaughter in Yorkshir, near,\\nA monstrous bastards, and do it contrary\\nThan pity men, with heigh! three your time trembles,\\nYou might have overhead on end: nay, stay,\\nTo tear up window himself. She's a woman\\nA torch-bear in blood!\\n\\nFirst Servingman:\\nWorthy keeps his hands?\\n\\nGLOUCESTER:\\nLady, come, my caught! requires none with him;\\nAnd all the wallars gave up the dread bright\\nThat always us to woo redoiced.\\n\\nThird Citizen:\\nNot away: be'll come thee dead; that they are so in halfour\\nfound this deech, yet, if he\\nbe a prisoner.\\n\\nANGELO:\\nHark ye;\\nThanks, pray, some blusters; and we hear her for the\\nfire:--for it\\nPlucket this wedched bath must strike thee mine,\\nTo let owe peril of this affection,\\nA dump enough to bear half married once;\\nFrom which affecting your ancestor's spread,\\nWhich may common men puts you there something\\ngood agreemby.\\n\\nCORIOLANUS:\\nThank you, good father; be brief.\\nYou that shall please thee for thy moth\"\n"," b\"ROMEO:\\nIs't, no! no.\\nBy heaven, of Buckingham's brame,\\nIf not pity man! a bleaded mirts,\\nHe breathed such evils, and be of joy,\\nThat ever heard she was broke in dagger.\\nIn blessed grandam true oracles,\\nTravelligent again outrage, drums, and safe, she's dance;\\nFear not the heavens have done; thou runn'st us not.\\n\\nKING HENRY VI:\\nFarewell, my conscience in the time. What\\nmeans himself?\\n\\nPETRUCHIO:\\nAnd takes himself longer, I accept your business,\\nAnd being good, many, and smmit me foo\\nlessers than by consuming but the steely phosolet; or\\npale deadly home, shoetious-like too,\\nWhich in my sword and safety of royal teeth;\\nHither straight; for this ighorn lies heavy banish'd,\\nAs 'twere to go about dine and the tenter\\nSend us, pray heaven after dishonours' service.\\n\\nWARWICK:\\nI take my enemy; and so I was anointed:\\nSince thou, spetious island slander,--\\n\\nJULIET:\\nOur prayers do so, I swear it, and\\ntruly.\\n\\nLUCI:\\nOnce more, son-thy parties of a king:\\nBut, Warwick, till; and so, my warp-gentled,\\nAn o'era\"\n"," b\"ROMEO:\\nThe ditten pays; which is ere for\\nsome ill planing.\\nThough thou set'st a king,--\\nFalse--with my soldiers, ladies, wanton,\\nThat thought o' the fire, and now, away!\\nThy father gave of heaven! I will make me\\nattain the cunning of this place. But what, is not to over\\nThe phesent death hath lost it were,\\nWhich prompted of them, Is wept my heart!\\nI'll pay thy part, earth ease to Fortunal,\\nShall I redress thy beauty to death thou hast most kiss,\\nDesign'd my person all that must not seem.\\nWill you speak gone ill, yet his heart to see?\\nI lie, and bid, time's fond of love;\\nFetch in this royal philoies, and when it was\\nOve days in the house of Lancaster.\\n\\nKING HENRY VI:\\nBe pitch'd, and go before the gentlewon,\\nAnd may not, sopter than a mad-take tripute\\nEnsh'd in Baptista take; give me already.\\nWithin this maid is safe?\\n\\nKING HENRY VI:\\nAnd lay the lord protector all, of flesh,\\nBut sevil cannot undestain; and fighting\\nSo many things that bred a medal, state\\nWith one flower about the sea-side of a\"], shape=(5,), dtype=string) \n","\n","________________________________________________________________________________\n","\n","Run time: 2.933893918991089\n"]}],"source":["start = time.time()\n","states = None\n","next_char = tf.constant(['ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:'])\n","result = [next_char]\n","\n","for n in range(1000):\n","  next_char, states = one_step_model.generate_one_step(next_char, states=states)\n","  result.append(next_char)\n","\n","result = tf.strings.join(result)\n","end = time.time()\n","print(result, '\\n\\n' + '_'*80)\n","print('\\nRun time:', end - start)"]},{"cell_type":"markdown","id":"a98ef85f","metadata":{"papermill":{"duration":0.161497,"end_time":"2023-07-14T16:35:00.118744","exception":false,"start_time":"2023-07-14T16:34:59.957247","status":"completed"},"tags":[]},"source":["# 7. Export the generator\n","<div id=\"Export\"></div>"]},{"cell_type":"markdown","id":"694989ca","metadata":{"papermill":{"duration":0.165931,"end_time":"2023-07-14T16:35:00.447129","exception":false,"start_time":"2023-07-14T16:35:00.281198","status":"completed"},"tags":[]},"source":["This single-step model can easily be [saved and restored](https://www.tensorflow.org/guide/saved_model), allowing you to use it anywhere a `tf.saved_model` is accepted."]},{"cell_type":"code","execution_count":42,"id":"e13d6210","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:00.771573Z","iopub.status.busy":"2023-07-14T16:35:00.771206Z","iopub.status.idle":"2023-07-14T16:35:07.529663Z","shell.execute_reply":"2023-07-14T16:35:07.528686Z"},"papermill":{"duration":6.923342,"end_time":"2023-07-14T16:35:07.532187","exception":false,"start_time":"2023-07-14T16:35:00.608845","status":"completed"},"tags":[]},"outputs":[],"source":["tf.saved_model.save(one_step_model, 'one_step')\n","one_step_reloaded = tf.saved_model.load('one_step')"]},{"cell_type":"code","execution_count":43,"id":"160fe65a","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:07.854565Z","iopub.status.busy":"2023-07-14T16:35:07.854213Z","iopub.status.idle":"2023-07-14T16:35:08.393793Z","shell.execute_reply":"2023-07-14T16:35:08.392639Z"},"papermill":{"duration":0.703716,"end_time":"2023-07-14T16:35:08.396024","exception":false,"start_time":"2023-07-14T16:35:07.692308","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["ROMEO:\n","Is there move, by some sentence a coast\n","Ah sortething both these great power hours seal on\n","the joyf\n"]}],"source":["states = None\n","next_char = tf.constant(['ROMEO:'])\n","result = [next_char]\n","\n","for n in range(100):\n","  next_char, states = one_step_reloaded.generate_one_step(next_char, states=states)\n","  result.append(next_char)\n","\n","print(tf.strings.join(result)[0].numpy().decode(\"utf-8\"))"]},{"cell_type":"markdown","id":"2f3d1bd2","metadata":{"papermill":{"duration":0.159141,"end_time":"2023-07-14T16:35:08.715906","exception":false,"start_time":"2023-07-14T16:35:08.556765","status":"completed"},"tags":[]},"source":["# 8. Advanced: Customized Training\n","<div id=\"Advance\"></div>"]},{"cell_type":"markdown","id":"7fcf5ed6","metadata":{"papermill":{"duration":0.170718,"end_time":"2023-07-14T16:35:09.047358","exception":false,"start_time":"2023-07-14T16:35:08.87664","status":"completed"},"tags":[]},"source":["The above training procedure is simple, but does not give you much control.\n","It uses teacher-forcing which prevents bad predictions from being fed back to the model, so the model never learns to recover from mistakes.\n","\n","So now that you've seen how to run the model manually next you'll implement the training loop. This gives a starting point if, for example, you want to implement _curriculum  learning_ to help stabilize the model's open-loop output.\n","\n","The most important part of a custom training loop is the train step function.\n","\n","Use `tf.GradientTape` to track the gradients. You can learn more about this approach by reading the [eager execution guide](https://www.tensorflow.org/guide/eager).\n","\n","The basic procedure is:\n","\n","1. Execute the model and calculate the loss under a `tf.GradientTape`.\n","2. Calculate the updates and apply them to the model using the optimizer."]},{"cell_type":"code","execution_count":44,"id":"753d85df","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:09.369102Z","iopub.status.busy":"2023-07-14T16:35:09.368714Z","iopub.status.idle":"2023-07-14T16:35:09.375529Z","shell.execute_reply":"2023-07-14T16:35:09.374483Z"},"papermill":{"duration":0.171943,"end_time":"2023-07-14T16:35:09.377938","exception":false,"start_time":"2023-07-14T16:35:09.205995","status":"completed"},"tags":[]},"outputs":[],"source":["class CustomTraining(MyModel):\n","  @tf.function\n","  def train_step(self, inputs):\n","      inputs, labels = inputs\n","      with tf.GradientTape() as tape:\n","          predictions = self(inputs, training=True)\n","          loss = self.loss(labels, predictions)\n","      grads = tape.gradient(loss, model.trainable_variables)\n","      self.optimizer.apply_gradients(zip(grads, model.trainable_variables))\n","\n","      return {'loss': loss}"]},{"cell_type":"markdown","id":"fc0d1ec5","metadata":{"papermill":{"duration":0.157477,"end_time":"2023-07-14T16:35:09.694113","exception":false,"start_time":"2023-07-14T16:35:09.536636","status":"completed"},"tags":[]},"source":["The above implementation of the train_step method follows Keras' train_step conventions. This is optional, but it allows you to change the behavior of the train step and still use keras' Model.compile and Model.fit methods."]},{"cell_type":"code","execution_count":45,"id":"7259b30b","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:10.016577Z","iopub.status.busy":"2023-07-14T16:35:10.015843Z","iopub.status.idle":"2023-07-14T16:35:10.030474Z","shell.execute_reply":"2023-07-14T16:35:10.029468Z"},"papermill":{"duration":0.17984,"end_time":"2023-07-14T16:35:10.032541","exception":false,"start_time":"2023-07-14T16:35:09.852701","status":"completed"},"tags":[]},"outputs":[],"source":["model = CustomTraining(\n","    vocab_size=len(ids_from_chars.get_vocabulary()),\n","    embedding_dim=embedding_dim,\n","    rnn_units=rnn_units)"]},{"cell_type":"code","execution_count":46,"id":"4b582333","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:10.354539Z","iopub.status.busy":"2023-07-14T16:35:10.354171Z","iopub.status.idle":"2023-07-14T16:35:10.365999Z","shell.execute_reply":"2023-07-14T16:35:10.365136Z"},"papermill":{"duration":0.175155,"end_time":"2023-07-14T16:35:10.367922","exception":false,"start_time":"2023-07-14T16:35:10.192767","status":"completed"},"tags":[]},"outputs":[],"source":["model.compile(optimizer = tf.keras.optimizers.Adam(),\n","              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True))"]},{"cell_type":"code","execution_count":47,"id":"2aa9537b","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:35:10.687853Z","iopub.status.busy":"2023-07-14T16:35:10.687479Z","iopub.status.idle":"2023-07-14T16:35:22.988883Z","shell.execute_reply":"2023-07-14T16:35:22.987908Z"},"papermill":{"duration":12.464362,"end_time":"2023-07-14T16:35:22.991038","exception":false,"start_time":"2023-07-14T16:35:10.526676","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["172/172 [==============================] - 12s 44ms/step - loss: 2.7349\n"]},{"data":{"text/plain":["<keras.callbacks.History at 0x7ae9667a76d0>"]},"execution_count":47,"metadata":{},"output_type":"execute_result"}],"source":["model.fit(dataset, epochs=1)"]},{"cell_type":"markdown","id":"99f4b28b","metadata":{"papermill":{"duration":0.164837,"end_time":"2023-07-14T16:35:23.32189","exception":false,"start_time":"2023-07-14T16:35:23.157053","status":"completed"},"tags":[]},"source":["Or if you need more control, you can write your own complete custom training loop:"]},{"cell_type":"code","execution_count":48,"id":"625a355f","metadata":{"_kg_hide-output":true,"execution":{"iopub.execute_input":"2023-07-14T16:35:23.656395Z","iopub.status.busy":"2023-07-14T16:35:23.65602Z","iopub.status.idle":"2023-07-14T16:36:49.672297Z","shell.execute_reply":"2023-07-14T16:36:49.670733Z"},"papermill":{"duration":86.184657,"end_time":"2023-07-14T16:36:49.674522","exception":false,"start_time":"2023-07-14T16:35:23.489865","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["Epoch 1 Batch 0 Loss 2.1678\n","Epoch 1 Batch 50 Loss 2.0802\n","Epoch 1 Batch 100 Loss 2.0112\n","Epoch 1 Batch 150 Loss 1.8511\n","\n","Epoch 1 Loss: 2.0027\n","Time taken for 1 epoch 10.41 sec\n","________________________________________________________________________________\n","Epoch 2 Batch 0 Loss 1.8497\n","Epoch 2 Batch 50 Loss 1.7764\n","Epoch 2 Batch 100 Loss 1.7243\n","Epoch 2 Batch 150 Loss 1.6311\n","\n","Epoch 2 Loss: 1.7264\n","Time taken for 1 epoch 8.40 sec\n","________________________________________________________________________________\n","Epoch 3 Batch 0 Loss 1.6037\n","Epoch 3 Batch 50 Loss 1.6014\n","Epoch 3 Batch 100 Loss 1.5823\n","Epoch 3 Batch 150 Loss 1.5163\n","\n","Epoch 3 Loss: 1.5627\n","Time taken for 1 epoch 8.56 sec\n","________________________________________________________________________________\n","Epoch 4 Batch 0 Loss 1.4605\n","Epoch 4 Batch 50 Loss 1.4537\n","Epoch 4 Batch 100 Loss 1.4121\n","Epoch 4 Batch 150 Loss 1.4463\n","\n","Epoch 4 Loss: 1.4620\n","Time taken for 1 epoch 8.41 sec\n","________________________________________________________________________________\n","Epoch 5 Batch 0 Loss 1.3981\n","Epoch 5 Batch 50 Loss 1.3998\n","Epoch 5 Batch 100 Loss 1.3724\n","Epoch 5 Batch 150 Loss 1.3425\n","\n","Epoch 5 Loss: 1.3926\n","Time taken for 1 epoch 8.44 sec\n","________________________________________________________________________________\n","Epoch 6 Batch 0 Loss 1.2927\n","Epoch 6 Batch 50 Loss 1.3384\n","Epoch 6 Batch 100 Loss 1.3421\n","Epoch 6 Batch 150 Loss 1.3360\n","\n","Epoch 6 Loss: 1.3402\n","Time taken for 1 epoch 8.24 sec\n","________________________________________________________________________________\n","Epoch 7 Batch 0 Loss 1.2872\n","Epoch 7 Batch 50 Loss 1.2613\n","Epoch 7 Batch 100 Loss 1.2822\n","Epoch 7 Batch 150 Loss 1.2924\n","\n","Epoch 7 Loss: 1.2955\n","Time taken for 1 epoch 8.39 sec\n","________________________________________________________________________________\n","Epoch 8 Batch 0 Loss 1.2136\n","Epoch 8 Batch 50 Loss 1.2509\n","Epoch 8 Batch 100 Loss 1.2421\n","Epoch 8 Batch 150 Loss 1.2321\n","\n","Epoch 8 Loss: 1.2547\n","Time taken for 1 epoch 8.29 sec\n","________________________________________________________________________________\n","Epoch 9 Batch 0 Loss 1.1800\n","Epoch 9 Batch 50 Loss 1.1956\n","Epoch 9 Batch 100 Loss 1.2157\n","Epoch 9 Batch 150 Loss 1.2257\n","\n","Epoch 9 Loss: 1.2161\n","Time taken for 1 epoch 8.34 sec\n","________________________________________________________________________________\n","Epoch 10 Batch 0 Loss 1.1294\n","Epoch 10 Batch 50 Loss 1.1662\n","Epoch 10 Batch 100 Loss 1.2165\n","Epoch 10 Batch 150 Loss 1.2160\n","\n","Epoch 10 Loss: 1.1788\n","Time taken for 1 epoch 8.39 sec\n","________________________________________________________________________________\n"]}],"source":["EPOCHS = 10\n","\n","mean = tf.metrics.Mean()\n","\n","for epoch in range(EPOCHS):\n","    start = time.time()\n","\n","    mean.reset_states()\n","    for (batch_n, (inp, target)) in enumerate(dataset):\n","        logs = model.train_step([inp, target])\n","        mean.update_state(logs['loss'])\n","\n","        if batch_n % 50 == 0:\n","            template = f\"Epoch {epoch+1} Batch {batch_n} Loss {logs['loss']:.4f}\"\n","            print(template)\n","\n","    # saving (checkpoint) the model every 5 epochs\n","    if (epoch + 1) % 5 == 0:\n","        model.save_weights(checkpoint_prefix.format(epoch=epoch))\n","\n","    print()\n","    print(f'Epoch {epoch+1} Loss: {mean.result().numpy():.4f}')\n","    print(f'Time taken for 1 epoch {time.time() - start:.2f} sec')\n","    print(\"_\"*80)\n","\n","model.save_weights(checkpoint_prefix.format(epoch=epoch))"]},{"cell_type":"code","execution_count":49,"id":"8de67bab","metadata":{"execution":{"iopub.execute_input":"2023-07-14T16:36:50.018927Z","iopub.status.busy":"2023-07-14T16:36:50.018563Z","iopub.status.idle":"2023-07-14T16:36:53.502831Z","shell.execute_reply":"2023-07-14T16:36:53.50191Z"},"papermill":{"duration":3.658931,"end_time":"2023-07-14T16:36:53.505354","exception":false,"start_time":"2023-07-14T16:36:49.846423","status":"completed"},"tags":[]},"outputs":[{"name":"stdout","output_type":"stream","text":["[<tf.Tensor: shape=(5,), dtype=string, numpy=\n","array([b'ROMEO:', b'ROMEO:', b'ROMEO:', b'ROMEO:', b'ROMEO:'],\n","      dtype=object)>]\n","tf.Tensor([b'\\n' b'\\n' b'\\n' b'\\n' b'\\n'], shape=(5,), dtype=string)\n","tf.Tensor([b'I' b'W' b'M' b'N' b'N'], shape=(5,), dtype=string)\n","tf.Tensor([b' ' b'h' b'y' b'o' b'o'], shape=(5,), dtype=string)\n","\n","\n","________________________________________________________________________________\n","[<tf.Tensor: shape=(5,), dtype=string, numpy=\n","array([b'ROMEO:', b'ROMEO:', b'ROMEO:', b'ROMEO:', b'ROMEO:'],\n","      dtype=object)>, <tf.Tensor: shape=(5,), dtype=string, numpy=array([b'\\n', b'\\n', b'\\n', b'\\n', b'\\n'], dtype=object)>, <tf.Tensor: shape=(5,), dtype=string, numpy=array([b'I', b'W', b'M', b'N', b'N'], dtype=object)>, <tf.Tensor: shape=(5,), dtype=string, numpy=array([b' ', b'h', b'y', b'o', b'o'], dtype=object)>, <tf.Tensor: shape=(5,), dtype=string, numpy=array([b'h', b'e', b' ', b'w', b'w'], dtype=object)>]\n","\n","\n","________________________________________________________________________________\n","tf.Tensor(\n","[b\"ROMEO:\\nI have rid me, the five true wealth me.\\n\\nNurse:\\nHow now, silence?\\n\\nMUCIII:\\nI can require anchor!\\n\\nProvost:\\nGod save the trate.\\n\\nShepherd:\\nAlack, he hate must say?\\n\\nJULIET:\\nO Romeo, the best? I will have to death\\nCase in the sweet with youth.\\n\\nISABELLA:\\nI am a good words, and then in wrong'd,\\nAntonious grace must invenlished: be it suffice\\nto you.\\n\\nFirst Murderer:\\nAlan, we talk here doth purge the instruct me\\njusts and after him.\\n\\nFirst Musician:\\nWhat, Joon, fair cannot hear a thanking-sceigh.\\nIf that thus nature is he hath carried one intend:\\nNor I sust not nime as this.\\n\\nLUCIO:\\nIn warrant for whose consent? where is Romeo\\nHeary hearted in three; for it hath mine have\\nAre clant to they are never lodged on a man of men in vain,\\nAnd, if not rear his myet height.\\nI'll give thee that\\nThou speak'st that strange with a dream,\\nAnd whose pretting ill be my inhal no loss.\\n\\nKATHARINA:\\nThen wits he be unding: because I die!\\n\\nRIVERS:\\nAnd there you loved the heavens look still and ask'd.\\n\\nShepherd\"\n"," b\"ROMEO:\\nWhere is that?\\n\\nSecond Murderer:\\nThen, to the horse that; saying the wide?\\nDoth the title, contrarament.\\nTime hence, be gone! and so II't we aslect it.\\n\\nDUKE OF YORK:\\nWhere he is his argued! I will make his person; have;\\nFor 'tis an honest man, what of love we have leave to do\\nI spegh of him.\\n\\nSecond Murderer:\\nI will, if I less than come to't?\\n\\nMENENIUS:\\nYet with answer in whose wounds me fall, farewell.\\n\\nKATHARINA:\\nCome, come, my lord, so art, my life, a centremens at\\nVostering into the enemy, fawness in Rons.\\n\\nKATHARIIA:\\nI prithee, nicht.\\n\\nDUKE OF YORK:\\nProda,! You account thy firsh more complaint\\nTo chide by mine honour,\\nOr from the wretch hath done not, for, sir, who\\ncouldst thou hath he territ you and the sun.\\nI would fall locks let thee with a sorterer:\\nGrod his belong their inlections. Therefore lady!\\n\\nKATHARINA:\\nWith whom she can curms this night?\\n\\nROMEO:\\nAy.\\n\\nGONZALO:\\nHis taught,\\nWhen I fors, shall findness it; look his brother,\\nOr if you call you: help me to thy honour,\\nI wa\"\n"," b\"ROMEO:\\nMy rage is best, in an admirance,\\nTho woo me to the use that affords.\\n\\nKING EDWARD IV:\\n\\nKING RICHARD III:\\nIf she were bound to the cause?\\n\\nMAPTIUS:\\nHe's beats,\\nI pray, prettory the grace my brother's, sir,\\nTake on common judgment hine, or no.\\n\\nESCALUS:\\nCome hither, warrial, he\\nhath contribt'd my Edward steel\\nIn what you would wash. He is a went to\\nwoo! what a plot hath coming how had whore,\\nTranslaft kipe the father of the ears.\\n\\nBRUTUS:\\nWho shall remose you that must die the\\nbeggo forbid! She might the king withs and tuil'd Eavio.\\n\\nDUKE VINCENTIO:\\nA fight of him that I wash the prison! hear may he down\\nevers an honey and easy till no more hand,\\nAnd first they say: it is so answer the bay;\\nThou detain'd his good, this to meet her either'd,\\nTo quest myself.\\n\\nBALTTAS:\\nStits your numberable, very father, whose bare against\\nThe childness of my song and saves. Or field,\\nI am continuence; but if that be at\\npursue yields.\\n\\nKING LEdIUS:\\nShall we sew thy.\\n\\nMENENIUS:\\nWhat shelkes country he eve\"\n"," b\"ROMEO:\\nNow, by your brother's blood to you,\\nFrom five anofted form, sir? why, Volsces whom:\\nLoved me resion; sir, the lords behope the end,\\nWhich gently traction waste to one fear\\nMy conscience tapping for the josom.\\n\\nNORTHUMBERLAND:\\nMy wager it undertake to know the city.\\n\\nPAgase:\\nI shath\\nmiscived to a hisper? O heavy live\\nTouching the seas' to meet what thou hast made here,\\nHelp fall of wife: take leave of this youngh.\\nCome, Petruchio! Here it is humour.\\n\\nSLY:\\nI warrant the black reproace than the shade you'ld.\\n\\nLUCIO:\\nNay, by my woff; and therefore beseech\\nWith a city parts so increase:' they prokes\\nApartly as 't withdraw than your heart's coups\\nYierse thing; therein thou from the saint,\\nAnd send you on: if it be Poxt gently\\nThan He should be laugh to home: 'Perdorant\\nHere craSsing Hereford by Edward's wife: the capless vestal came\\nEre thou hadst thus can nurse to the wall to say:\\nAnd that the seas with nothing creed.\\nGood nurse, commend me with false foot\\nWas not to request yourselves hi\"\n"," b\"ROMEO:\\nNow, not for God's cries of my confenety.\\n\\nROMEO:\\nAy, am I, a goddesh, sir; it is,\\nWe ento thee; if he is false well-faced\\nAnd send them, the matter, by the expless indeed\\nNay, not the shepherd's thost:\\nNor Edward night the shoulder deeds.\\nI think, is he intend visages: neither amen!\\nAnd take it, forward, be plaint with some woman,\\nAnd see his rage\\nbreast thine eyes, and to the most? Hark you;\\nHer witch is not in most grief of the\\nThirdy sent better far and enjoy:\\nBesides he dids!\\n\\nFirst Lord:\\nNow thou that have misfod me to speak.\\n\\nNORTHUMBERLAND:\\nHaste she would already, and he took for\\nAnd kiss 'em; yes, ere thy dark-orifucasion.\\n\\nchoto all once,\\nTo fiery friend with agree-dib;\\nBe paltiful trate, a degree of noble keep.\\nGive she breathe death, and am not by sadwel,\\nI am as one of bloody. As my tresh, for your read.\\nThen he hath lived his queen the issue\\nWhat I may have this mounned hine,\\nI know the lips that but such a place;\\nGod an ungots till I am to his name strange?\\n\\nLEONTES:\\nI\"], shape=(5,), dtype=string) \n","\n","________________________________________________________________________________\n","\n","Run time: 3.4399213790893555\n"]}],"source":["one_step_model = OneStep(model, chars_from_ids, ids_from_chars)\n","\n","start = time.time()\n","\n","\n","states = None\n","next_char = tf.constant(['ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:'])\n","result = [next_char]\n","print(result)\n","i = 0\n","for n in range(1000):\n","  next_char, states = one_step_model.generate_one_step(next_char, states=states)\n","  if( i<3):\n","    print(next_char)\n","  i+=1\n","  result.append(next_char)\n","  if(i==4):\n","    print('\\n\\n' + '_'*80)\n","    print(result)\n","    print('\\n\\n' + '_'*80)\n","\n","result = tf.strings.join(result)\n","end = time.time()\n","print(result, '\\n\\n' + '_'*80)\n","print('\\nRun time:', end - start)"]},{"cell_type":"markdown","id":"6d1a05f1","metadata":{"papermill":{"duration":0.172441,"end_time":"2023-07-14T16:36:53.897615","exception":false,"start_time":"2023-07-14T16:36:53.725174","status":"completed"},"tags":[]},"source":["<div class=\"alert alert-block alert-success\" style=\"font-size:30px\">\n","    <b>Thank you for coming to the end 🎉</b>\n","    <br>If it was useful to you, then I will be glad to see your vote.<br>\n","    </div>"]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.10.12"},"papermill":{"default_parameters":{},"duration":341.916553,"end_time":"2023-07-14T16:36:56.927161","environment_variables":{},"exception":null,"input_path":"__notebook__.ipynb","output_path":"__notebook__.ipynb","parameters":{},"start_time":"2023-07-14T16:31:15.010608","version":"2.4.0"}},"nbformat":4,"nbformat_minor":5}