TV Script Generation

In this project, you'll generate your own Simpsons TV scripts using RNNs. You'll be using part of the Simpsons dataset of scripts from 27 seasons. The Neural Network you'll build will generate a new TV script for a scene at Moe's Tavern.

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..


In [8]:
"""
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 data
text = text[81:]

Explore the Data

Play around with view_sentence_range to view different parts of the data.


In [6]:
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 Stats
Roughly the number of unique words: 11492
Number of scenes: 262
Average number of sentences in each scene: 15.248091603053435
Number of lines: 4257
Average 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)


In [14]:
import numpy as np
import problem_unittests as tests

from collections import Counter

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
    cnt = Counter()
    for word in text:
        cnt[word] += 1
    vocab_to_int = {}
    int_to_vocab = {}
    for i, (word, count) in enumerate(cnt.most_common()):
        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 Passed

Tokenize 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||".


In [15]:
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
    token_dict = {'.': '||period||',
                  ',': '||comma||',
                  '"': '||quotation_mark||',
                  ';': '||semicolon||',
                  '!': '||exclamation_mark||',
                  '?': '||question_mark||',
                  '(': '||left_parentheses||',
                  ')': '||right_parentheses||',
                  '--': '||dash||',
                  '\n': '||return|'}
    return token_dict

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_tokenize(token_lookup)


Tests Passed

Preprocess all the data and save it

Running the code cell below will preprocess all the data and save it to file.


In [16]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Preprocess Training, Validation, and Testing Data
helper.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.


In [9]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
import numpy as np
import 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


In [10]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer'
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if 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: 1.0.0
Default GPU Device: /gpu:0

Input

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 name parameter.
  • Targets placeholder
  • Learning Rate placeholder

Return the placeholders in the following the tuple (Input, Targets, LearingRate)


In [11]:
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, shape=[None, None], name='input')
    targets = tf.placeholder(tf.int32, shape=[None, None], name='targets')
    learning_rate = tf.placeholder(tf.float32, name='learning_rate')
    return input, targets, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_inputs(get_inputs)


Tests Passed

Build 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()

Return the cell and initial state in the following tuple (Cell, InitialState)


In [12]:
lstm_layers = 2

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)
    # TODO: The script generation script doesn't have the capability to set keep_prob to 1.
    #drop = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob)
    cell = tf.contrib.rnn.MultiRNNCell([lstm] * lstm_layers)
    initial_state = cell.zero_state(batch_size, tf.float32)
    initial_state = tf.identity(initial_state, name='initial_state')
    return cell, initial_state


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_init_cell(get_init_cell)


Tests Passed

Word Embedding

Apply embedding to input_data using TensorFlow. Return the embedded sequence.


In [13]:
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
    embedding = tf.Variable(tf.random_uniform([vocab_size, embed_dim], minval=-1, maxval=1))
    embed = tf.nn.embedding_lookup(embedding, input_data)
    return embed


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_embed(get_embed)


Tests Passed

Build RNN

You created a RNN Cell in the get_init_cell() function. Time to use the cell to create a RNN.

Return the outputs and final_state state in the following tuple (Outputs, FinalState)


In [14]:
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, final_state = tf.nn.dynamic_rnn(cell, inputs, dtype=tf.float32)
    final_state = tf.identity(final_state, name='final_state')
    return outputs, final_state


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_build_rnn(build_rnn)


Tests Passed

Build the Neural Network

Apply the functions you implemented above to:

  • Apply embedding to input_data using your get_embed(input_data, vocab_size, embed_dim) function.
  • Build RNN using cell and your build_rnn(cell, inputs) function.
  • Apply a fully connected layer with a linear activation and vocab_size as the number of outputs.

Return the logits and final state in the following tuple (Logits, FinalState)


In [15]:
embed_dim = 300

def build_nn(cell, rnn_size, input_data, vocab_size):
    """
    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
    :return: Tuple (Logits, FinalState)
    """
    # TODO: Implement Function
    embed = get_embed(input_data, vocab_size, embed_dim)
    outputs, final_state = build_rnn(cell, embed)
    logits = tf.contrib.layers.fully_connected(outputs, vocab_size, activation_fn=None)
    return logits, final_state


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_build_nn(build_nn)


Tests Passed

Batches

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 exmple, get_batches([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], 2, 3) would return a Numpy array of the following:

[
  # First Batch
  [
    # Batch of Input
    [[ 1  2  3], [ 7  8  9]],
    # Batch of targets
    [[ 2  3  4], [ 8  9 10]]
  ],

  # Second Batch
  [
    # Batch of Input
    [[ 4  5  6], [10 11 12]],
    # Batch of targets
    [[ 5  6  7], [11 12 13]]
  ]
]

In [16]:
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
    batch = batch_size*seq_length
    n_batches = (len(int_text) - 1)//batch
    int_text = int_text[:n_batches*batch + 1]
    
    batches = np.zeros((n_batches, 2, batch_size, seq_length), dtype=np.int32)
    
    for i in range(0, n_batches):
        for j in range(0, batch_size):
            idx = (j*n_batches + i)*seq_length
            batches[i][0][j] = int_text[idx:idx+seq_length]
            batches[i][1][j] = int_text[idx+1:idx+seq_length+1]
    return batches


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_batches(get_batches)


Tests Passed

Neural Network Training

Hyperparameters

Tune the following parameters:

  • Set num_epochs to the number of epochs.
  • Set batch_size to the batch size.
  • Set rnn_size to the size of the RNNs.
  • Set seq_length to the length of sequence.
  • Set learning_rate to the learning rate.
  • Set show_every_n_batches to the number of batches the neural network should print progress.

In [17]:
# Number of Epochs
num_epochs = 500
# Batch Size
batch_size = 256
# RNN Size
rnn_size = 512
# Sequence Length
seq_length = 30
# Learning Rate
learning_rate = 0.001
# Show stats for every n number of batches
show_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.


In [18]:
"""
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)

    # 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]
    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 forms to see if anyone is having the same problem.


In [ ]:
"""
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    0/8   train_loss = 8.820
Epoch   2 Batch    0/8   train_loss = 6.149
Epoch   4 Batch    0/8   train_loss = 6.007
Epoch   6 Batch    0/8   train_loss = 5.971
Epoch   8 Batch    0/8   train_loss = 5.952
Epoch  10 Batch    0/8   train_loss = 5.935
Epoch  12 Batch    0/8   train_loss = 5.888
Epoch  14 Batch    0/8   train_loss = 5.812
Epoch  16 Batch    0/8   train_loss = 5.739
Epoch  18 Batch    0/8   train_loss = 5.672
Epoch  20 Batch    0/8   train_loss = 5.615
Epoch  22 Batch    0/8   train_loss = 5.565
Epoch  24 Batch    0/8   train_loss = 5.515
Epoch  26 Batch    0/8   train_loss = 5.463
Epoch  28 Batch    0/8   train_loss = 5.403
Epoch  30 Batch    0/8   train_loss = 5.339
Epoch  32 Batch    0/8   train_loss = 5.269
Epoch  34 Batch    0/8   train_loss = 5.195
Epoch  36 Batch    0/8   train_loss = 5.125
Epoch  38 Batch    0/8   train_loss = 5.054
Epoch  40 Batch    0/8   train_loss = 4.983
Epoch  42 Batch    0/8   train_loss = 4.915
Epoch  44 Batch    0/8   train_loss = 4.849
Epoch  46 Batch    0/8   train_loss = 4.791
Epoch  48 Batch    0/8   train_loss = 4.743
Epoch  50 Batch    0/8   train_loss = 4.682
Epoch  52 Batch    0/8   train_loss = 4.624
Epoch  54 Batch    0/8   train_loss = 4.572
Epoch  56 Batch    0/8   train_loss = 4.526
Epoch  58 Batch    0/8   train_loss = 4.489
Epoch  60 Batch    0/8   train_loss = 4.451
Epoch  62 Batch    0/8   train_loss = 4.403
Epoch  64 Batch    0/8   train_loss = 4.355
Epoch  66 Batch    0/8   train_loss = 4.313
Epoch  68 Batch    0/8   train_loss = 4.272
Epoch  70 Batch    0/8   train_loss = 4.246
Epoch  72 Batch    0/8   train_loss = 4.209
Epoch  74 Batch    0/8   train_loss = 4.168
Epoch  76 Batch    0/8   train_loss = 4.131
Epoch  78 Batch    0/8   train_loss = 4.102
Epoch  80 Batch    0/8   train_loss = 4.076
Epoch  82 Batch    0/8   train_loss = 4.035
Epoch  84 Batch    0/8   train_loss = 4.012
Epoch  86 Batch    0/8   train_loss = 3.980
Epoch  88 Batch    0/8   train_loss = 3.965
Epoch  90 Batch    0/8   train_loss = 3.987
Epoch  92 Batch    0/8   train_loss = 4.038
Epoch  94 Batch    0/8   train_loss = 3.962
Epoch  96 Batch    0/8   train_loss = 3.891
Epoch  98 Batch    0/8   train_loss = 3.820
Epoch 100 Batch    0/8   train_loss = 3.822
Epoch 102 Batch    0/8   train_loss = 3.816
Epoch 104 Batch    0/8   train_loss = 3.787
Epoch 106 Batch    0/8   train_loss = 3.705
Epoch 108 Batch    0/8   train_loss = 3.675
Epoch 110 Batch    0/8   train_loss = 3.651
Epoch 112 Batch    0/8   train_loss = 3.617
Epoch 114 Batch    0/8   train_loss = 3.636
Epoch 116 Batch    0/8   train_loss = 3.600
Epoch 118 Batch    0/8   train_loss = 3.522
Epoch 120 Batch    0/8   train_loss = 3.502
Epoch 122 Batch    0/8   train_loss = 3.472
Epoch 124 Batch    0/8   train_loss = 3.448
Epoch 126 Batch    0/8   train_loss = 3.404
Epoch 128 Batch    0/8   train_loss = 3.373
Epoch 130 Batch    0/8   train_loss = 3.330
Epoch 132 Batch    0/8   train_loss = 3.292
Epoch 134 Batch    0/8   train_loss = 3.252
Epoch 136 Batch    0/8   train_loss = 3.265
Epoch 138 Batch    0/8   train_loss = 3.245
Epoch 140 Batch    0/8   train_loss = 3.191
Epoch 142 Batch    0/8   train_loss = 3.119
Epoch 144 Batch    0/8   train_loss = 3.081
Epoch 146 Batch    0/8   train_loss = 3.046
Epoch 148 Batch    0/8   train_loss = 3.093
Epoch 150 Batch    0/8   train_loss = 2.987
Epoch 152 Batch    0/8   train_loss = 2.989
Epoch 154 Batch    0/8   train_loss = 2.912
Epoch 156 Batch    0/8   train_loss = 2.879
Epoch 158 Batch    0/8   train_loss = 2.827
Epoch 160 Batch    0/8   train_loss = 2.774
Epoch 162 Batch    0/8   train_loss = 2.753
Epoch 164 Batch    0/8   train_loss = 2.710
Epoch 166 Batch    0/8   train_loss = 2.642
Epoch 168 Batch    0/8   train_loss = 2.614
Epoch 170 Batch    0/8   train_loss = 2.572
Epoch 172 Batch    0/8   train_loss = 2.568
Epoch 174 Batch    0/8   train_loss = 2.500
Epoch 176 Batch    0/8   train_loss = 2.481
Epoch 178 Batch    0/8   train_loss = 2.511
Epoch 180 Batch    0/8   train_loss = 2.461
Epoch 182 Batch    0/8   train_loss = 2.389
Epoch 184 Batch    0/8   train_loss = 2.323
Epoch 186 Batch    0/8   train_loss = 2.264
Epoch 188 Batch    0/8   train_loss = 2.237
Epoch 190 Batch    0/8   train_loss = 2.183
Epoch 192 Batch    0/8   train_loss = 2.186
Epoch 194 Batch    0/8   train_loss = 2.205
Epoch 196 Batch    0/8   train_loss = 2.117
Epoch 198 Batch    0/8   train_loss = 2.117
Epoch 200 Batch    0/8   train_loss = 2.019
Epoch 202 Batch    0/8   train_loss = 1.999
Epoch 204 Batch    0/8   train_loss = 1.932
Epoch 206 Batch    0/8   train_loss = 1.927
Epoch 208 Batch    0/8   train_loss = 1.888
Epoch 210 Batch    0/8   train_loss = 1.851
Epoch 212 Batch    0/8   train_loss = 1.814
Epoch 214 Batch    0/8   train_loss = 1.778
Epoch 216 Batch    0/8   train_loss = 1.745
Epoch 218 Batch    0/8   train_loss = 1.728
Epoch 220 Batch    0/8   train_loss = 1.667
Epoch 222 Batch    0/8   train_loss = 1.622
Epoch 224 Batch    0/8   train_loss = 1.598
Epoch 226 Batch    0/8   train_loss = 1.569
Epoch 228 Batch    0/8   train_loss = 1.539
Epoch 230 Batch    0/8   train_loss = 1.554
Epoch 232 Batch    0/8   train_loss = 1.516
Epoch 234 Batch    0/8   train_loss = 1.572
Epoch 236 Batch    0/8   train_loss = 1.503
Epoch 238 Batch    0/8   train_loss = 1.539
Epoch 240 Batch    0/8   train_loss = 1.391
Epoch 242 Batch    0/8   train_loss = 1.363
Epoch 244 Batch    0/8   train_loss = 1.308
Epoch 246 Batch    0/8   train_loss = 1.310
Epoch 248 Batch    0/8   train_loss = 1.270
Epoch 250 Batch    0/8   train_loss = 1.282
Epoch 252 Batch    0/8   train_loss = 1.216
Epoch 254 Batch    0/8   train_loss = 1.216
Epoch 256 Batch    0/8   train_loss = 1.189
Epoch 258 Batch    0/8   train_loss = 1.133
Epoch 260 Batch    0/8   train_loss = 1.109
Epoch 262 Batch    0/8   train_loss = 1.104
Epoch 264 Batch    0/8   train_loss = 1.082
Epoch 266 Batch    0/8   train_loss = 1.091
Epoch 268 Batch    0/8   train_loss = 1.172
Epoch 270 Batch    0/8   train_loss = 1.176
Epoch 272 Batch    0/8   train_loss = 1.018
Epoch 274 Batch    0/8   train_loss = 0.989
Epoch 276 Batch    0/8   train_loss = 0.945
Epoch 278 Batch    0/8   train_loss = 0.932
Epoch 280 Batch    0/8   train_loss = 0.914
Epoch 316 Batch    0/8   train_loss = 0.611
Epoch 318 Batch    0/8   train_loss = 0.596
Epoch 320 Batch    0/8   train_loss = 0.572
Epoch 322 Batch    0/8   train_loss = 0.567
Epoch 324 Batch    0/8   train_loss = 0.548
Epoch 326 Batch    0/8   train_loss = 0.560
Epoch 328 Batch    0/8   train_loss = 0.562
Epoch 330 Batch    0/8   train_loss = 0.592
Epoch 332 Batch    0/8   train_loss = 0.567
Epoch 334 Batch    0/8   train_loss = 0.535
Epoch 336 Batch    0/8   train_loss = 0.512
Epoch 338 Batch    0/8   train_loss = 0.468
Epoch 340 Batch    0/8   train_loss = 0.457
Epoch 342 Batch    0/8   train_loss = 0.447
Epoch 344 Batch    0/8   train_loss = 0.437
Epoch 346 Batch    0/8   train_loss = 0.431
Epoch 348 Batch    0/8   train_loss = 0.421
Epoch 350 Batch    0/8   train_loss = 0.418
Epoch 352 Batch    0/8   train_loss = 0.416
Epoch 354 Batch    0/8   train_loss = 0.420
Epoch 356 Batch    0/8   train_loss = 0.420
Epoch 358 Batch    0/8   train_loss = 0.390
Epoch 360 Batch    0/8   train_loss = 0.374
Epoch 362 Batch    0/8   train_loss = 0.354
Epoch 364 Batch    0/8   train_loss = 0.343
Epoch 366 Batch    0/8   train_loss = 0.333
Epoch 368 Batch    0/8   train_loss = 0.332
Epoch 370 Batch    0/8   train_loss = 0.319
Epoch 372 Batch    0/8   train_loss = 0.323
Epoch 374 Batch    0/8   train_loss = 0.323
Epoch 376 Batch    0/8   train_loss = 0.321
Epoch 378 Batch    0/8   train_loss = 0.314
Epoch 380 Batch    0/8   train_loss = 0.322
Epoch 382 Batch    0/8   train_loss = 0.309
Epoch 384 Batch    0/8   train_loss = 0.303
Epoch 386 Batch    0/8   train_loss = 0.287
Epoch 388 Batch    0/8   train_loss = 0.285
Epoch 390 Batch    0/8   train_loss = 0.263
Epoch 392 Batch    0/8   train_loss = 0.260
Epoch 394 Batch    0/8   train_loss = 0.250
Epoch 396 Batch    0/8   train_loss = 0.243
Epoch 398 Batch    0/8   train_loss = 0.245
Epoch 400 Batch    0/8   train_loss = 0.235
Epoch 402 Batch    0/8   train_loss = 0.228
Epoch 404 Batch    0/8   train_loss = 0.233
Epoch 406 Batch    0/8   train_loss = 0.229
Epoch 408 Batch    0/8   train_loss = 0.221
Epoch 410 Batch    0/8   train_loss = 0.216
Epoch 412 Batch    0/8   train_loss = 0.208
Epoch 414 Batch    0/8   train_loss = 0.204
Epoch 416 Batch    0/8   train_loss = 0.201
Epoch 418 Batch    0/8   train_loss = 0.198
Epoch 420 Batch    0/8   train_loss = 0.195
Epoch 422 Batch    0/8   train_loss = 0.195
Epoch 424 Batch    0/8   train_loss = 0.196
Epoch 426 Batch    0/8   train_loss = 0.192
Epoch 428 Batch    0/8   train_loss = 0.186
Epoch 430 Batch    0/8   train_loss = 0.186
Epoch 432 Batch    0/8   train_loss = 0.183
Epoch 434 Batch    0/8   train_loss = 0.176
Epoch 436 Batch    0/8   train_loss = 0.172
Epoch 438 Batch    0/8   train_loss = 0.171
Epoch 440 Batch    0/8   train_loss = 0.168
Epoch 442 Batch    0/8   train_loss = 0.166
Epoch 444 Batch    0/8   train_loss = 0.164
Epoch 446 Batch    0/8   train_loss = 0.167
Epoch 448 Batch    0/8   train_loss = 0.165
Epoch 450 Batch    0/8   train_loss = 0.162
Epoch 452 Batch    0/8   train_loss = 0.166
Epoch 454 Batch    0/8   train_loss = 0.164
Epoch 456 Batch    0/8   train_loss = 0.162
Epoch 458 Batch    0/8   train_loss = 0.154
Epoch 460 Batch    0/8   train_loss = 0.150
Epoch 462 Batch    0/8   train_loss = 0.146
Epoch 464 Batch    0/8   train_loss = 0.143
Epoch 466 Batch    0/8   train_loss = 0.141
Epoch 468 Batch    0/8   train_loss = 0.140
Epoch 470 Batch    0/8   train_loss = 0.139
Epoch 472 Batch    0/8   train_loss = 0.138
Epoch 474 Batch    0/8   train_loss = 0.137
Epoch 476 Batch    0/8   train_loss = 0.135
Epoch 478 Batch    0/8   train_loss = 0.134
Epoch 480 Batch    0/8   train_loss = 0.134
Epoch 482 Batch    0/8   train_loss = 0.133
Epoch 484 Batch    0/8   train_loss = 0.131
Epoch 486 Batch    0/8   train_loss = 0.129
Epoch 488 Batch    0/8   train_loss = 0.128
Epoch 490 Batch    0/8   train_loss = 0.127
Epoch 492 Batch    0/8   train_loss = 0.126
Epoch 494 Batch    0/8   train_loss = 0.126
Epoch 496 Batch    0/8   train_loss = 0.125
Epoch 498 Batch    0/8   train_loss = 0.123
Model Trained and Saved

lstm_layers = 1 batch_size = 256 rnn_size = 512 train_loss = 0.726 (200 epochs)

lstm_layers = 2 batch_size = 256 rnn_size = 512 train_loss = 2.163 (200 epochs), 0.123 (500 epochs)

Save Parameters

Save seq_length and save_dir for generating a new TV script.


In [20]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Save parameters for checkpoint
helper.save_params((seq_length, save_dir))

Checkpoint


In [21]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import tensorflow as tf
import numpy as np
import helper
import 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)


In [22]:
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
    input = loaded_graph.get_tensor_by_name('input:0')
    initial_state = loaded_graph.get_tensor_by_name('initial_state:0')
    final_state = loaded_graph.get_tensor_by_name('final_state:0')
    probabilities = loaded_graph.get_tensor_by_name('probs:0')
    return input, initial_state, final_state, probabilities


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_tensors(get_tensors)


Tests Passed

Choose Word

Implement the pick_word() function to select the next word using probabilities.


In [23]:
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
    vocab_size = len(int_to_vocab)
    return int_to_vocab[np.random.choice(vocab_size, 1, p=probabilities)[0]]


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_pick_word(pick_word)


Tests Passed

Generate TV Script

This will generate the TV script for you. Set gen_length to the length of TV script you want to generate.


In [24]:
gen_length = 200
# homer_simpson, moe_szyslak, or Barney_Gumble
prime_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[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)


moe_szyslak: ah, business is slow. people today are healthier and drinking less. you know, if it wasn't for the junior high school next door no one was i.
homer_simpson: i am, moe?
moe_szyslak:(more) you know. i just open it?
moe_szyslak: okay, my stars... your friends challenge you all the kids...
homer_simpson: well, i guess you know more, bart?
homer_simpson: can't this place are you? get a grandkids car--(to moe) ooh.
homer_simpson: i can't wait all be a beer with my first place?
homer_simpson: no, you've got to the face?
lenny_leonard: i wanna be sitting to all of that duff, but you guess my car.
kent_brockman: uh, that's so i'd, noosey come on? then now i daddy, i couldn't be thought and just pick on the face who makes a problem in the chinese, moe.
lenny_leonard: who found it. everything i ever need to take my wife.
carl_carlson: i'm here

The TV Script is Nonsensical

It's ok if the TV script doesn't make any sense. We trained on less than a megabyte of text. In order to get good results, you'll have to use a smaller vocabulary or get more data. Luckly there's more data! As we mentioned in the begging of this project, this is a subset of another dataset. We didn't have you train on all the data, because that would take too long. However, you are free to train your neural network on all the data. After you complete the project, of course.

Submitting This Project

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_tv_script_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.