TV script generation using deep learning
In this project, a Recurrent Neural Networks is used to generate a TV script for a scene at Moe’s Tavern using some of the data from Simpsons dataset of scripts from 27 seasons. Here is the code for this project which was a part of Udacity’s Deep Learning Nanodegree.
Get the Data
The data is already provided for you. You’ll be using a subset of the original dataset. It consists of only the scenes in Moe’s Tavern. This doesn’t include other versions of the tavern, like “Moe’s Cavern”, “Flaming Moe’s”, “Uncle Moe’s Family Feed-Bag”, etc..
"""DON'T MODIFY ANYTHING IN THIS CELL"""import helper
data_dir = './data/simpsons/moes_tavern_lines.txt'text = helper.load_data(data_dir)# Ignore notice, since we don't use it for analysing the datatext = text[81:]# print(text)Moe_Szyslak: (INTO PHONE) Moe's Tavern. Where the elite meet to drink.Bart_Simpson: Eh, yeah, hello, is Mike there? Last name, Rotch.Moe_Szyslak: (INTO PHONE) Hold on, I'll check. (TO BARFLIES) Mike Rotch. Mike Rotch. Hey, has anybody seen Mike Rotch, lately?Moe_Szyslak: (INTO PHONE) Listen you little puke. One of these days I'm gonna catch you, and I'm gonna carve my name on your back with an ice pick.Moe_Szyslak: What's the matter Homer? You're not your normal effervescent self.Homer_Simpson: I got my problems, Moe. Give me another one.Moe_Szyslak: Homer, hey, you should not drink to forget your problems.Barney_Gumble: Yeah, you should only drink to enhance your social skills.Explore the Data
Play around with view_sentence_range to view different parts of the data.
view_sentence_range = (0, 10)
"""DON'T MODIFY ANYTHING IN THIS CELL"""import numpy as np
print('Dataset Stats')print('Roughly the number of unique words: {}'.format(len({word: None for word in text.split()})))scenes = text.split('\n\n')print('Number of scenes: {}'.format(len(scenes)))sentence_count_scene = [scene.count('\n') for scene in scenes]print('Average number of sentences in each scene: {}'.format(np.average(sentence_count_scene)))
sentences = [sentence for scene in scenes for sentence in scene.split('\n')]print('Number of lines: {}'.format(len(sentences)))word_count_sentence = [len(sentence.split()) for sentence in sentences]print('Average number of words in each line: {}'.format(np.average(word_count_sentence)))
print()print('The sentences {} to {}:'.format(*view_sentence_range))print('\n'.join(text.split('\n')[view_sentence_range[0]:view_sentence_range[1]]))Dataset StatsRoughly the number of unique words: 11492Number of scenes: 262Average number of sentences in each scene: 15.248091603053435Number of lines: 4257Average number of words in each line: 11.50434578341555
The sentences 0 to 10:Moe_Szyslak: (INTO PHONE) Moe's Tavern. Where the elite meet to drink.Bart_Simpson: Eh, yeah, hello, is Mike there? Last name, Rotch.Moe_Szyslak: (INTO PHONE) Hold on, I'll check. (TO BARFLIES) Mike Rotch. Mike Rotch. Hey, has anybody seen Mike Rotch, lately?Moe_Szyslak: (INTO PHONE) Listen you little puke. One of these days I'm gonna catch you, and I'm gonna carve my name on your back with an ice pick.Moe_Szyslak: What's the matter Homer? You're not your normal effervescent self.Homer_Simpson: I got my problems, Moe. Give me another one.Moe_Szyslak: Homer, hey, you should not drink to forget your problems.Barney_Gumble: Yeah, you should only drink to enhance your social skills.Implement Preprocessing Functions
The first thing to do to any dataset is preprocessing. Implement the following preprocessing functions below:
- Lookup Table
- Tokenize Punctuation
Lookup Table
To create a word embedding, you first need to transform the words to ids. In this function, create two dictionaries:
- Dictionary to go from the words to an id, we’ll call
vocab_to_int - Dictionary to go from the id to word, we’ll call
int_to_vocab
Return these dictionaries in the following tuple (vocab_to_int, int_to_vocab)
import numpy as npimport problem_unittests as tests
def create_lookup_tables(text): """ Create lookup tables for vocabulary :param text: The text of tv scripts split into words :return: A tuple of dicts (vocab_to_int, int_to_vocab) """ # TODO: Implement Function # print(text[:10]) unique_words = set(text) # print (unique_words)
vocab_to_int = {} int_to_vocab = {}
for i, word in enumerate(unique_words): vocab_to_int[word] = i int_to_vocab[i] = word
return (vocab_to_int, int_to_vocab)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_create_lookup_tables(create_lookup_tables)Tests PassedTokenize Punctuation
We’ll be splitting the script into a word array using spaces as delimiters. However, punctuations like periods and exclamation marks make it hard for the neural network to distinguish between the word “bye” and “bye!“.
Implement the function token_lookup to return a dict that will be used to tokenize symbols like ”!” into “||Exclamation_Mark||“. Create a dictionary for the following symbols where the symbol is the key and value is the token:
- Period ( . )
- Comma ( , )
- Quotation Mark ( ” )
- Semicolon ( ; )
- Exclamation mark ( ! )
- Question mark ( ? )
- Left Parentheses ( ( )
- Right Parentheses ( ) )
- Dash ( — )
- Return ( \n )
This dictionary will be used to token the symbols and add the delimiter (space) around it. This separates the symbols as it’s own word, making it easier for the neural network to predict on the next word. Make sure you don’t use a token that could be confused as a word. Instead of using the token “dash”, try using something like “||dash||“.
def token_lookup(): """ Generate a dict to turn punctuation into a token. :return: Tokenize dictionary where the key is the punctuation and the value is the token """ # TODO: Implement Function tokens_dict = { '.': '||period||', ',': '||comma||', '"': '||quotation_mark||', ';': '||semicolon||', '!': '||exclamation_mark||', '?': '||question_mark||', '(': '||left_parentheses||', ')': '||right_parentheses||', '--': '||dash||', '\n': '||return||' } return tokens_dict
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_tokenize(token_lookup)Tests PassedPreprocess all the data and save it
Running the code cell below will preprocess all the data and save it to file.
"""DON'T MODIFY ANYTHING IN THIS CELL"""# Preprocess Training, Validation, and Testing Datahelper.preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)Check Point
This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk.
"""DON'T MODIFY ANYTHING IN THIS CELL"""import helperimport numpy as npimport problem_unittests as tests
int_text, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()Build the Neural Network
You’ll build the components necessary to build a RNN by implementing the following functions below:
- get_inputs
- get_init_cell
- get_embed
- build_rnn
- build_nn
- get_batches
Check the Version of TensorFlow and Access to GPU
"""DON'T MODIFY ANYTHING IN THIS CELL"""from distutils.version import LooseVersionimport warningsimport tensorflow as tf
# Check TensorFlow Versionassert LooseVersion(tf.__version__) >= LooseVersion('1.3'), 'Please use TensorFlow version 1.3 or newer'print('TensorFlow Version: {}'.format(tf.__version__))
# Check for a GPUif not tf.test.gpu_device_name(): warnings.warn('No GPU found. Please use a GPU to train your neural network.')else: print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))TensorFlow Version: 1.3.0Default GPU Device: /gpu:0Input
Implement the get_inputs() function to create TF Placeholders for the Neural Network. It should create the following placeholders:
- Input text placeholder named “input” using the TF Placeholder
nameparameter. - Targets placeholder
- Learning Rate placeholder
Return the placeholders in the following tuple (Input, Targets, LearningRate)
import numpy as npimport problem_unittests as tests
def get_inputs(): """ Create TF Placeholders for input, targets, and learning rate. :return: Tuple (input, targets, learning rate) """ # TODO: Implement Function Input = tf.placeholder(tf.int32, [None, None], name = "input") Targets = tf.placeholder(tf.int32, [None, None], name = "targets") LearningRate = tf.placeholder(tf.float32, name = "learning_rate") return (Input, Targets, LearningRate)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_get_inputs(get_inputs)Tests PassedBuild RNN Cell and Initialize
Stack one or more BasicLSTMCells in a MultiRNNCell.
- The Rnn size should be set using
rnn_size - Initalize Cell State using the MultiRNNCell’s
zero_state()function- Apply the name “initial_state” to the initial state using
tf.identity()
- Apply the name “initial_state” to the initial state using
Return the cell and initial state in the following tuple (Cell, InitialState)
def get_init_cell(batch_size, rnn_size): """ Create an RNN Cell and initialize it. :param batch_size: Size of batches :param rnn_size: Size of RNNs :return: Tuple (cell, initialize state) """ # TODO: Implement Function lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size) Cell = tf.contrib.rnn.MultiRNNCell([lstm] * 3) InitialState = Cell.zero_state(batch_size, tf.float32) InitialState = tf.identity(InitialState, name="initial_state") return (Cell, InitialState)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_get_init_cell(get_init_cell)Tests PassedWord Embedding
Apply embedding to input_data using TensorFlow. Return the embedded sequence.
def get_embed(input_data, vocab_size, embed_dim): """ Create embedding for <input_data>. :param input_data: TF placeholder for text input. :param vocab_size: Number of words in vocabulary. :param embed_dim: Number of embedding dimensions :return: Embedded input. """ # TODO: Implement Function # https://www.tensorflow.org/api_docs/python/tf/contrib/layers/embed_sequence return tf.contrib.layers.embed_sequence(input_data, vocab_size, embed_dim)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_get_embed(get_embed)Tests PassedBuild RNN
You created a RNN Cell in the get_init_cell() function. Time to use the cell to create a RNN.
- Build the RNN using the
tf.nn.dynamic_rnn() - Apply the name “final_state” to the final state using
tf.identity()
Return the outputs and final_state state in the following tuple (Outputs, FinalState)
def build_rnn(cell, inputs): """ Create a RNN using a RNN Cell :param cell: RNN Cell :param inputs: Input text data :return: Tuple (Outputs, Final State) """ # TODO: Implement Function (Outputs, FinalState) = tf.nn.dynamic_rnn(cell, inputs, dtype=tf.float32) FinalState = tf.identity(FinalState, name="final_state") return (Outputs, FinalState)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_build_rnn(build_rnn)Tests PassedBuild the Neural Network
Apply the functions you implemented above to:
- Apply embedding to
input_datausing yourget_embed(input_data, vocab_size, embed_dim)function. - Build RNN using
celland yourbuild_rnn(cell, inputs)function. - Apply a fully connected layer with a linear activation and
vocab_sizeas the number of outputs.
Return the logits and final state in the following tuple (Logits, FinalState)
def build_nn(cell, rnn_size, input_data, vocab_size, embed_dim): """ Build part of the neural network :param cell: RNN cell :param rnn_size: Size of rnns :param input_data: Input data :param vocab_size: Vocabulary size :param embed_dim: Number of embedding dimensions :return: Tuple (Logits, FinalState) """ # TODO: Implement Function embedded_input = get_embed(input_data, vocab_size, embed_dim) rnn_output, FinalState = build_rnn(cell, embedded_input) Logits = tf.contrib.layers.fully_connected(rnn_output, vocab_size, activation_fn=None) return (Logits, FinalState)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_build_nn(build_nn)Tests PassedBatches
Implement get_batches to create batches of input and targets using int_text. The batches should be a Numpy array with the shape (number of batches, 2, batch size, sequence length). Each batch contains two elements:
- The first element is a single batch of input with the shape
[batch size, sequence length] - The second element is a single batch of targets with the shape
[batch size, sequence length]
If you can’t fill the last batch with enough data, drop the last batch.
For example, get_batches([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], 3, 2) would return a Numpy array of the following:
[ # First Batch [ # Batch of Input [[ 1 2], [ 7 8], [13 14]] # Batch of targets [[ 2 3], [ 8 9], [14 15]] ]
# Second Batch [ # Batch of Input [[ 3 4], [ 9 10], [15 16]] # Batch of targets [[ 4 5], [10 11], [16 17]] ]
# Third Batch [ # Batch of Input [[ 5 6], [11 12], [17 18]] # Batch of targets [[ 6 7], [12 13], [18 1]] ]]Notice that the last target value in the last batch is the first input value of the first batch. In this case, 1. This is a common technique used when creating sequence batches, although it is rather unintuitive.
def get_batches(int_text, batch_size, seq_length): """ Return batches of input and target :param int_text: Text with the words replaced by their ids :param batch_size: The size of batch :param seq_length: The length of sequence :return: Batches as a Numpy array """ # TODO: Implement Function inputs_per_batch = batch_size * seq_length num_batches = len(int_text)//(inputs_per_batch) int_text = int_text[:num_batches*inputs_per_batch] # drop unused int_text.append(int_text[0]) # to use first input value of first batch as last target value of the last batch
# allocate memory with shape of batches batches = np.zeros([num_batches, 2, batch_size, seq_length], dtype=np.int32)
# Add seq_length elements at a time to input and targets appropriately for i in range(0, len(int_text), seq_length): batch_no = (i // seq_length) % num_batches index_in_batch = i // (seq_length * num_batches)
if (index_in_batch == batch_size): break
# input batches[batch_no, 0, index_in_batch] = int_text[i : i+seq_length]
# targets batches[batch_no, 1, index_in_batch] = int_text[i+1 : i+seq_length+1] # element next to input element return batches
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_get_batches(get_batches)Tests PassedNeural Network Training
Hyperparameters
Tune the following parameters:
- Set
num_epochsto the number of epochs. - Set
batch_sizeto the batch size. - Set
rnn_sizeto the size of the RNNs. - Set
embed_dimto the size of the embedding. - Set
seq_lengthto the length of sequence. - Set
learning_rateto the learning rate. - Set
show_every_n_batchesto the number of batches the neural network should print progress.
# Number of Epochsnum_epochs = 300 # loss was still going down after 128 epochs but started increasing after 400# Batch Sizebatch_size = 128# RNN Sizernn_size = 256# Embedding Dimension Sizeembed_dim = 256# Sequence Lengthseq_length = 32# Learning Ratelearning_rate = 0.01# Show stats for every n number of batchesshow_every_n_batches = 16
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""save_dir = './save'Build the Graph
Build the graph using the neural network you implemented.
"""DON'T MODIFY ANYTHING IN THIS CELL"""from tensorflow.contrib import seq2seq
train_graph = tf.Graph()with train_graph.as_default(): vocab_size = len(int_to_vocab) input_text, targets, lr = get_inputs() input_data_shape = tf.shape(input_text) cell, initial_state = get_init_cell(input_data_shape[0], rnn_size) logits, final_state = build_nn(cell, rnn_size, input_text, vocab_size, embed_dim)
# Probabilities for generating words probs = tf.nn.softmax(logits, name='probs')
# Loss function cost = seq2seq.sequence_loss( logits, targets, tf.ones([input_data_shape[0], input_data_shape[1]]))
# Optimizer optimizer = tf.train.AdamOptimizer(lr)
# Gradient Clipping gradients = optimizer.compute_gradients(cost) capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients if grad is not None] train_op = optimizer.apply_gradients(capped_gradients)Train
Train the neural network on the preprocessed data. If you have a hard time getting a good loss, check the forums to see if anyone is having the same problem.
"""DON'T MODIFY ANYTHING IN THIS CELL"""batches = get_batches(int_text, batch_size, seq_length)
with tf.Session(graph=train_graph) as sess: sess.run(tf.global_variables_initializer())
for epoch_i in range(num_epochs): state = sess.run(initial_state, {input_text: batches[0][0]})
for batch_i, (x, y) in enumerate(batches): feed = { input_text: x, targets: y, initial_state: state, lr: learning_rate} train_loss, state, _ = sess.run([cost, final_state, train_op], feed)
# Show every <show_every_n_batches> batches if (epoch_i * len(batches) + batch_i) % show_every_n_batches == 0: print('Epoch {:>3} Batch {:>4}/{} train_loss = {:.3f}'.format( epoch_i, batch_i, len(batches), train_loss))
# Save Model saver = tf.train.Saver() saver.save(sess, save_dir) print('Model Trained and Saved')Epoch 0 Batch 0/16 train_loss = 8.822Epoch 1 Batch 0/16 train_loss = 6.426Epoch 2 Batch 0/16 train_loss = 6.098Epoch 3 Batch 0/16 train_loss = 6.074Epoch 4 Batch 0/16 train_loss = 6.044Epoch 5 Batch 0/16 train_loss = 6.026Epoch 6 Batch 0/16 train_loss = 6.027Epoch 7 Batch 0/16 train_loss = 6.010Epoch 8 Batch 0/16 train_loss = 6.010Epoch 9 Batch 0/16 train_loss = 5.997Epoch 10 Batch 0/16 train_loss = 6.012Epoch 11 Batch 0/16 train_loss = 5.989Epoch 12 Batch 0/16 train_loss = 5.978Epoch 13 Batch 0/16 train_loss = 5.980Epoch 14 Batch 0/16 train_loss = 5.938Epoch 15 Batch 0/16 train_loss = 5.910Epoch 16 Batch 0/16 train_loss = 5.890Epoch 17 Batch 0/16 train_loss = 5.867Epoch 18 Batch 0/16 train_loss = 5.854Epoch 19 Batch 0/16 train_loss = 5.784Epoch 20 Batch 0/16 train_loss = 5.343Epoch 21 Batch 0/16 train_loss = 5.027Epoch 22 Batch 0/16 train_loss = 4.800Epoch 23 Batch 0/16 train_loss = 4.605Epoch 24 Batch 0/16 train_loss = 4.449Epoch 25 Batch 0/16 train_loss = 4.327Epoch 26 Batch 0/16 train_loss = 4.179Epoch 27 Batch 0/16 train_loss = 4.064Epoch 28 Batch 0/16 train_loss = 3.952Epoch 29 Batch 0/16 train_loss = 3.842Epoch 30 Batch 0/16 train_loss = 3.767Epoch 31 Batch 0/16 train_loss = 3.766Epoch 32 Batch 0/16 train_loss = 3.623Epoch 33 Batch 0/16 train_loss = 3.474Epoch 34 Batch 0/16 train_loss = 3.370Epoch 35 Batch 0/16 train_loss = 3.328Epoch 36 Batch 0/16 train_loss = 3.213Epoch 37 Batch 0/16 train_loss = 3.185Epoch 38 Batch 0/16 train_loss = 3.078Epoch 39 Batch 0/16 train_loss = 2.988Epoch 40 Batch 0/16 train_loss = 2.943Epoch 41 Batch 0/16 train_loss = 2.858Epoch 42 Batch 0/16 train_loss = 2.773Epoch 43 Batch 0/16 train_loss = 2.738Epoch 44 Batch 0/16 train_loss = 2.723Epoch 45 Batch 0/16 train_loss = 2.734Epoch 46 Batch 0/16 train_loss = 2.683Epoch 47 Batch 0/16 train_loss = 2.592Epoch 48 Batch 0/16 train_loss = 2.543Epoch 49 Batch 0/16 train_loss = 2.474Epoch 50 Batch 0/16 train_loss = 2.364Epoch 51 Batch 0/16 train_loss = 2.271Epoch 52 Batch 0/16 train_loss = 2.196Epoch 53 Batch 0/16 train_loss = 2.148Epoch 54 Batch 0/16 train_loss = 2.119Epoch 55 Batch 0/16 train_loss = 2.098Epoch 56 Batch 0/16 train_loss = 2.115Epoch 57 Batch 0/16 train_loss = 2.040Epoch 58 Batch 0/16 train_loss = 1.962Epoch 59 Batch 0/16 train_loss = 1.930Epoch 60 Batch 0/16 train_loss = 1.847Epoch 61 Batch 0/16 train_loss = 1.756Epoch 62 Batch 0/16 train_loss = 1.742Epoch 63 Batch 0/16 train_loss = 1.694Epoch 64 Batch 0/16 train_loss = 1.637Epoch 65 Batch 0/16 train_loss = 1.597Epoch 66 Batch 0/16 train_loss = 1.513Epoch 67 Batch 0/16 train_loss = 1.441Epoch 68 Batch 0/16 train_loss = 1.363Epoch 69 Batch 0/16 train_loss = 1.348Epoch 70 Batch 0/16 train_loss = 1.293Epoch 71 Batch 0/16 train_loss = 1.288Epoch 72 Batch 0/16 train_loss = 1.312Epoch 73 Batch 0/16 train_loss = 1.286Epoch 74 Batch 0/16 train_loss = 1.234Epoch 75 Batch 0/16 train_loss = 1.210Epoch 76 Batch 0/16 train_loss = 1.223Epoch 77 Batch 0/16 train_loss = 1.164Epoch 78 Batch 0/16 train_loss = 1.123Epoch 79 Batch 0/16 train_loss = 1.069Epoch 80 Batch 0/16 train_loss = 1.016Epoch 81 Batch 0/16 train_loss = 0.991Epoch 82 Batch 0/16 train_loss = 0.946Epoch 83 Batch 0/16 train_loss = 0.899Epoch 84 Batch 0/16 train_loss = 0.873Epoch 85 Batch 0/16 train_loss = 0.875Epoch 86 Batch 0/16 train_loss = 0.860Epoch 87 Batch 0/16 train_loss = 0.804Epoch 88 Batch 0/16 train_loss = 0.812Epoch 89 Batch 0/16 train_loss = 0.817Epoch 90 Batch 0/16 train_loss = 0.784Epoch 91 Batch 0/16 train_loss = 0.741Epoch 92 Batch 0/16 train_loss = 0.706Epoch 93 Batch 0/16 train_loss = 0.698Epoch 94 Batch 0/16 train_loss = 0.693Epoch 95 Batch 0/16 train_loss = 0.667Epoch 96 Batch 0/16 train_loss = 0.671Epoch 97 Batch 0/16 train_loss = 0.647Epoch 98 Batch 0/16 train_loss = 0.694Epoch 99 Batch 0/16 train_loss = 0.708Epoch 100 Batch 0/16 train_loss = 0.721Epoch 101 Batch 0/16 train_loss = 0.756Epoch 102 Batch 0/16 train_loss = 0.754Epoch 103 Batch 0/16 train_loss = 0.739Epoch 104 Batch 0/16 train_loss = 0.806Epoch 105 Batch 0/16 train_loss = 0.909Epoch 106 Batch 0/16 train_loss = 0.967Epoch 107 Batch 0/16 train_loss = 0.987Epoch 108 Batch 0/16 train_loss = 0.941Epoch 109 Batch 0/16 train_loss = 0.856Epoch 110 Batch 0/16 train_loss = 0.838Epoch 111 Batch 0/16 train_loss = 0.815Epoch 112 Batch 0/16 train_loss = 0.769Epoch 113 Batch 0/16 train_loss = 0.719Epoch 114 Batch 0/16 train_loss = 0.671Epoch 115 Batch 0/16 train_loss = 0.653Epoch 116 Batch 0/16 train_loss = 0.630Epoch 117 Batch 0/16 train_loss = 0.637Epoch 118 Batch 0/16 train_loss = 0.612Epoch 119 Batch 0/16 train_loss = 0.546Epoch 120 Batch 0/16 train_loss = 0.482Epoch 121 Batch 0/16 train_loss = 0.478Epoch 122 Batch 0/16 train_loss = 0.449Epoch 123 Batch 0/16 train_loss = 0.442Epoch 124 Batch 0/16 train_loss = 0.413Epoch 125 Batch 0/16 train_loss = 0.376Epoch 126 Batch 0/16 train_loss = 0.351Epoch 127 Batch 0/16 train_loss = 0.344Epoch 128 Batch 0/16 train_loss = 0.331Epoch 129 Batch 0/16 train_loss = 0.298Epoch 130 Batch 0/16 train_loss = 0.297Epoch 131 Batch 0/16 train_loss = 0.263Epoch 132 Batch 0/16 train_loss = 0.223Epoch 133 Batch 0/16 train_loss = 0.227Epoch 134 Batch 0/16 train_loss = 0.221Epoch 135 Batch 0/16 train_loss = 0.205Epoch 136 Batch 0/16 train_loss = 0.208Epoch 137 Batch 0/16 train_loss = 0.190Epoch 138 Batch 0/16 train_loss = 0.172Epoch 139 Batch 0/16 train_loss = 0.171Epoch 140 Batch 0/16 train_loss = 0.195Epoch 141 Batch 0/16 train_loss = 0.172Epoch 142 Batch 0/16 train_loss = 0.157Epoch 143 Batch 0/16 train_loss = 0.152Epoch 144 Batch 0/16 train_loss = 0.140Epoch 145 Batch 0/16 train_loss = 0.133Epoch 146 Batch 0/16 train_loss = 0.128Epoch 147 Batch 0/16 train_loss = 0.124Epoch 148 Batch 0/16 train_loss = 0.121Epoch 149 Batch 0/16 train_loss = 0.117Epoch 150 Batch 0/16 train_loss = 0.116Epoch 151 Batch 0/16 train_loss = 0.114Epoch 152 Batch 0/16 train_loss = 0.112Epoch 153 Batch 0/16 train_loss = 0.111Epoch 154 Batch 0/16 train_loss = 0.109Epoch 155 Batch 0/16 train_loss = 0.108Epoch 156 Batch 0/16 train_loss = 0.107Epoch 157 Batch 0/16 train_loss = 0.106Epoch 158 Batch 0/16 train_loss = 0.104Epoch 159 Batch 0/16 train_loss = 0.104Epoch 160 Batch 0/16 train_loss = 0.103Epoch 161 Batch 0/16 train_loss = 0.102Epoch 162 Batch 0/16 train_loss = 0.101Epoch 163 Batch 0/16 train_loss = 0.101Epoch 164 Batch 0/16 train_loss = 0.100Epoch 165 Batch 0/16 train_loss = 0.099Epoch 166 Batch 0/16 train_loss = 0.099Epoch 167 Batch 0/16 train_loss = 0.098Epoch 168 Batch 0/16 train_loss = 0.097Epoch 169 Batch 0/16 train_loss = 0.097Epoch 170 Batch 0/16 train_loss = 0.096Epoch 171 Batch 0/16 train_loss = 0.096Epoch 172 Batch 0/16 train_loss = 0.095Epoch 173 Batch 0/16 train_loss = 0.095Epoch 174 Batch 0/16 train_loss = 0.094Epoch 175 Batch 0/16 train_loss = 0.094Epoch 176 Batch 0/16 train_loss = 0.094Epoch 177 Batch 0/16 train_loss = 0.093Epoch 178 Batch 0/16 train_loss = 0.093Epoch 179 Batch 0/16 train_loss = 0.092Epoch 180 Batch 0/16 train_loss = 0.092Epoch 181 Batch 0/16 train_loss = 0.092Epoch 182 Batch 0/16 train_loss = 0.091Epoch 183 Batch 0/16 train_loss = 0.091Epoch 184 Batch 0/16 train_loss = 0.091Epoch 185 Batch 0/16 train_loss = 0.091Epoch 186 Batch 0/16 train_loss = 0.090Epoch 187 Batch 0/16 train_loss = 0.090Epoch 188 Batch 0/16 train_loss = 0.090Epoch 189 Batch 0/16 train_loss = 0.090Epoch 190 Batch 0/16 train_loss = 0.089Epoch 191 Batch 0/16 train_loss = 0.089Epoch 192 Batch 0/16 train_loss = 0.089Epoch 193 Batch 0/16 train_loss = 0.089Epoch 194 Batch 0/16 train_loss = 0.088Epoch 195 Batch 0/16 train_loss = 0.088Epoch 196 Batch 0/16 train_loss = 0.088Epoch 197 Batch 0/16 train_loss = 0.088Epoch 198 Batch 0/16 train_loss = 0.088Epoch 199 Batch 0/16 train_loss = 0.087Epoch 200 Batch 0/16 train_loss = 0.087Epoch 201 Batch 0/16 train_loss = 0.087Epoch 202 Batch 0/16 train_loss = 0.087Epoch 203 Batch 0/16 train_loss = 0.087Epoch 204 Batch 0/16 train_loss = 0.087Epoch 205 Batch 0/16 train_loss = 0.086Epoch 206 Batch 0/16 train_loss = 0.086Epoch 207 Batch 0/16 train_loss = 0.086Epoch 208 Batch 0/16 train_loss = 0.086Epoch 209 Batch 0/16 train_loss = 0.086Epoch 210 Batch 0/16 train_loss = 0.086Epoch 211 Batch 0/16 train_loss = 0.086Epoch 212 Batch 0/16 train_loss = 0.086Epoch 213 Batch 0/16 train_loss = 0.085Epoch 214 Batch 0/16 train_loss = 0.085Epoch 215 Batch 0/16 train_loss = 0.085Epoch 216 Batch 0/16 train_loss = 0.085Epoch 217 Batch 0/16 train_loss = 0.085Epoch 218 Batch 0/16 train_loss = 0.085Epoch 219 Batch 0/16 train_loss = 0.085Epoch 220 Batch 0/16 train_loss = 0.085Epoch 221 Batch 0/16 train_loss = 0.085Epoch 222 Batch 0/16 train_loss = 0.085Epoch 223 Batch 0/16 train_loss = 0.084Epoch 224 Batch 0/16 train_loss = 0.084Epoch 225 Batch 0/16 train_loss = 0.084Epoch 226 Batch 0/16 train_loss = 0.084Epoch 227 Batch 0/16 train_loss = 0.084Epoch 228 Batch 0/16 train_loss = 0.084Epoch 229 Batch 0/16 train_loss = 0.085Epoch 230 Batch 0/16 train_loss = 0.090Epoch 231 Batch 0/16 train_loss = 0.100Epoch 232 Batch 0/16 train_loss = 0.172Epoch 233 Batch 0/16 train_loss = 0.723Epoch 234 Batch 0/16 train_loss = 2.116Epoch 235 Batch 0/16 train_loss = 2.848Epoch 236 Batch 0/16 train_loss = 2.818Epoch 237 Batch 0/16 train_loss = 2.644Epoch 238 Batch 0/16 train_loss = 2.422Epoch 239 Batch 0/16 train_loss = 2.131Epoch 240 Batch 0/16 train_loss = 1.869Epoch 241 Batch 0/16 train_loss = 1.657Epoch 242 Batch 0/16 train_loss = 1.537Epoch 243 Batch 0/16 train_loss = 1.358Epoch 244 Batch 0/16 train_loss = 1.259Epoch 245 Batch 0/16 train_loss = 1.167Epoch 246 Batch 0/16 train_loss = 1.078Epoch 247 Batch 0/16 train_loss = 0.996Epoch 248 Batch 0/16 train_loss = 0.936Epoch 249 Batch 0/16 train_loss = 0.884Epoch 250 Batch 0/16 train_loss = 0.866Epoch 251 Batch 0/16 train_loss = 0.817Epoch 252 Batch 0/16 train_loss = 0.760Epoch 253 Batch 0/16 train_loss = 0.720Epoch 254 Batch 0/16 train_loss = 0.682Epoch 255 Batch 0/16 train_loss = 0.643Epoch 256 Batch 0/16 train_loss = 0.626Epoch 257 Batch 0/16 train_loss = 0.596Epoch 258 Batch 0/16 train_loss = 0.571Epoch 259 Batch 0/16 train_loss = 0.537Epoch 260 Batch 0/16 train_loss = 0.534Epoch 261 Batch 0/16 train_loss = 0.517Epoch 262 Batch 0/16 train_loss = 0.501Epoch 263 Batch 0/16 train_loss = 0.478Epoch 264 Batch 0/16 train_loss = 0.487Epoch 265 Batch 0/16 train_loss = 0.510Epoch 266 Batch 0/16 train_loss = 0.488Epoch 267 Batch 0/16 train_loss = 0.463Epoch 268 Batch 0/16 train_loss = 0.494Epoch 269 Batch 0/16 train_loss = 0.479Epoch 270 Batch 0/16 train_loss = 0.485Epoch 271 Batch 0/16 train_loss = 0.474Epoch 272 Batch 0/16 train_loss = 0.469Epoch 273 Batch 0/16 train_loss = 0.463Epoch 274 Batch 0/16 train_loss = 0.455Epoch 275 Batch 0/16 train_loss = 0.429Epoch 276 Batch 0/16 train_loss = 0.435Epoch 277 Batch 0/16 train_loss = 0.471Epoch 278 Batch 0/16 train_loss = 0.460Epoch 279 Batch 0/16 train_loss = 0.483Epoch 280 Batch 0/16 train_loss = 0.448Epoch 281 Batch 0/16 train_loss = 0.395Epoch 282 Batch 0/16 train_loss = 0.375Epoch 283 Batch 0/16 train_loss = 0.377Epoch 284 Batch 0/16 train_loss = 0.367Epoch 285 Batch 0/16 train_loss = 0.350Epoch 286 Batch 0/16 train_loss = 0.317Epoch 287 Batch 0/16 train_loss = 0.296Epoch 288 Batch 0/16 train_loss = 0.269Epoch 289 Batch 0/16 train_loss = 0.260Epoch 290 Batch 0/16 train_loss = 0.249Epoch 291 Batch 0/16 train_loss = 0.217Epoch 292 Batch 0/16 train_loss = 0.224Epoch 293 Batch 0/16 train_loss = 0.201Epoch 294 Batch 0/16 train_loss = 0.198Epoch 295 Batch 0/16 train_loss = 0.222Epoch 296 Batch 0/16 train_loss = 0.240Epoch 297 Batch 0/16 train_loss = 0.265Epoch 298 Batch 0/16 train_loss = 0.266Epoch 299 Batch 0/16 train_loss = 0.249Model Trained and SavedSave Parameters
Save seq_length and save_dir for generating a new TV script.
"""DON'T MODIFY ANYTHING IN THIS CELL"""# Save parameters for checkpointhelper.save_params((seq_length, save_dir))Checkpoint
"""DON'T MODIFY ANYTHING IN THIS CELL"""import tensorflow as tfimport numpy as npimport helperimport problem_unittests as tests
_, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()seq_length, load_dir = helper.load_params()Implement Generate Functions
Get Tensors
Get tensors from loaded_graph using the function get_tensor_by_name(). Get the tensors using the following names:
- “input:0"
- "initial_state:0"
- "final_state:0"
- "probs:0”
Return the tensors in the following tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)
def get_tensors(loaded_graph): """ Get input, initial state, final state, and probabilities tensor from <loaded_graph> :param loaded_graph: TensorFlow graph loaded from file :return: Tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor) """ # TODO: Implement Function InputTensor = loaded_graph.get_tensor_by_name("input:0") InitialStateTensor = loaded_graph.get_tensor_by_name("initial_state:0") FinalStateTensor = loaded_graph.get_tensor_by_name("final_state:0") ProbsTensor = loaded_graph.get_tensor_by_name("probs:0") return (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_get_tensors(get_tensors)Tests PassedChoose Word
Implement the pick_word() function to select the next word using probabilities.
def pick_word(probabilities, int_to_vocab): """ Pick the next word in the generated text :param probabilities: Probabilites of the next word :param int_to_vocab: Dictionary of word ids as the keys and words as the values :return: String of the predicted word """ # TODO: Implement Function # Return word with highest probability return int_to_vocab[np.argmax(probabilities)]
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""tests.test_pick_word(pick_word)Tests PassedGenerate TV Script
This will generate the TV script for you. Set gen_length to the length of TV script you want to generate.
gen_length = 200# homer_simpson, moe_szyslak, or Barney_Gumbleprime_word = 'moe_szyslak'
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""loaded_graph = tf.Graph()with tf.Session(graph=loaded_graph) as sess: # Load saved model loader = tf.train.import_meta_graph(load_dir + '.meta') loader.restore(sess, load_dir)
# Get Tensors from loaded model input_text, initial_state, final_state, probs = get_tensors(loaded_graph)
# Sentences generation setup gen_sentences = [prime_word + ':'] prev_state = sess.run(initial_state, {input_text: np.array([[1]])})
# Generate sentences for n in range(gen_length): # Dynamic Input dyn_input = [[vocab_to_int[word] for word in gen_sentences[-seq_length:]]] dyn_seq_length = len(dyn_input[0])
# Get Prediction probabilities, prev_state = sess.run( [probs, final_state], {input_text: dyn_input, initial_state: prev_state})
pred_word = pick_word(probabilities[0][dyn_seq_length-1], int_to_vocab)
gen_sentences.append(pred_word)
# Remove tokens tv_script = ' '.join(gen_sentences) for key, token in token_dict.items(): ending = ' ' if key in ['\n', '(', '"'] else '' tv_script = tv_script.replace(' ' + token.lower(), key) tv_script = tv_script.replace('\n ', '\n') tv_script = tv_script.replace('( ', '(')
print(tv_script)INFO:tensorflow:Restoring parameters from ./savemoe_szyslak:(into phone) moe's tavern. where the elite meet to drink to see my camera.moe_szyslak: wow, this is all that?moe_szyslak: hey, there's it all the money, homer?homer_simpson:(big smile) hey, homer, i've got some of us.homer_simpson:(pats stomach) i'll cover mr. burns, i'm going to see you to go to the half way to turn out his truth.
moe_szyslak:(chuckles) hey, you got it, homer?moe_szyslak:(shocked) don't have to bet twenty dollars on a or here.moe_szyslak:(slightly confused chuckle) i've gotta go to go to myself.(à la improv comic) okay, which one back to go to my snake handler.moe_szyslak: marge, maggie, moe! that plank's only for duff porn from no return to save that are a little girl, moe?lucius:(confused) there where you can go of this, it, sir?homer_simpson:(drunk) oh, that'sThe TV Script is Nonsensical
While most of the generated TV script above doesn’t make any sense, this is expected because we trained on less than a megabyte of text. Using a smaller vocabulary or more data should produce better results.