TV Script Generation

In this project, we'll generate our own Simpsons TV scripts using RNNs. We'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

We'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 [1]:
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 [3]:
view_sentence_range = (20, 30)

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 20 to 30:

Moe_Szyslak: Looks like this is the end.
Barney_Gumble: That's all right. I couldn't have led a richer life.
Barney_Gumble: So the next time somebody tells you county folk are good, honest people, you can spit in their faces for me!
Lisa_Simpson: I will, Mr. Gumbel. But if you'll excuse me, I'm profiling my dad for the school paper. I thought it would be neat to follow him around for a day to see what makes him tick.
Barney_Gumble: Oh, that's sweet. I used to follow my dad to a lot of bars too. (BELCH)
Moe_Szyslak: Here you go. One beer, one chocolate milk.
Lisa_Simpson: Uh, excuse me, I have the chocolate milk.
Moe_Szyslak: Oh.
Moe_Szyslak: What's the matter, Homer? The depressin' effects of alcohol usually don't kick in 'til closing time.

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 [4]:
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)
    """
    # Implement Function
    word_counts = Counter(text)
    sorted_vocab = sorted(word_counts, key=word_counts.get, reverse=True)
    int_to_vocab = {index: text for index, text in enumerate(sorted_vocab)}
    vocab_to_int = {text: index for index, text in int_to_vocab.items()}

    return vocab_to_int, int_to_vocab

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 [5]:
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
    """
    # Implement Function
    token_dict = {
        '.': '<<period>>',
        ',': '<<comma>>',
        '"': '<<quotation_mark>>',
        ';': '<<semicolon>>',
        '!': '<<exclamation_mark>>',
        '?': '<<question_mark>>',
        '(': '<<left_parentheses>>',
        ')': '<<right_parentheses>>',
        '--': '<<dash>>',
        '\n': '<<return>>',
    }
    
    return token_dict

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 [6]:
# Preprocess Training, Validation, and Testing Data
helper.preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)

Check Point

This is our first checkpoint.


In [7]:
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

We'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 [8]:
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.

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


In [9]:
def get_inputs():
    """
    Create TF Placeholders for input, targets, and learning rate.
    :return: Tuple (input, targets, learning rate)
    """
    # Implement Function
    inputs = tf.placeholder(tf.int32, [None, None], name='input')
    targets = tf.placeholder(tf.int32, [None, None], name='targets')
    learning_rate = tf.placeholder(tf.float32, name='learning_rate')
    
    return inputs, targets, learning_rate

tests.test_get_inputs(get_inputs)


Tests Passed

Build RNN Cell and Initialize

Stack one or more BasicLSTMCells in a MultiRNNCell.

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


In [22]:
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)
    """
    # Implement Function
    lstm_layers = 2
    keep_prob = 0.5
    
    # Use a basic LSTM cell
    lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
    # Add dropout to the cell
    drop = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob)
    
    # Stack up multiple LSTM layers, for deep learning
    cell = tf.contrib.rnn.MultiRNNCell([drop] * lstm_layers)
    initial_state = tf.identity(cell.zero_state(batch_size, tf.float32), name="initial_state")
    
    return cell, initial_state

tests.test_get_init_cell(get_init_cell)


Tests Passed

Word Embedding

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


In [23]:
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.
    """
    # Implement Function
    #with graph.as_default():
    
    embedding = tf.Variable(tf.random_uniform((vocab_size, embed_dim), -1, 1))
    embed = tf.nn.embedding_lookup(embedding, input_data)
    
    return embed

tests.test_get_embed(get_embed)


Tests Passed

Build RNN

We 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 [24]:
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)
    """
    # Implement Function
    #embed_size = 100
    #vocab_size = 100
    #embed = get_embed(inputs, vocab_size, embed_size)
    #outputs, final_state = tf.nn.dynamic_rnn(cell, embed, initial_state=initial_state)
    
    outputs, state = tf.nn.dynamic_rnn(cell, inputs, dtype=tf.float32)
    final_state = tf.identity(state, name = 'final_state')
    
    return outputs, final_state

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 [25]:
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)
    """
    # Implement Function
    build_rnn_input = get_embed(input_data, vocab_size, rnn_size)
    outputs, final_state = build_rnn(cell, build_rnn_input)
    logits = tf.contrib.layers.fully_connected(outputs, vocab_size, activation_fn=None)
    
    return logits, final_state


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]

In [26]:
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
    """
    # Implement Function
    n_batches = int(len(int_text) / (batch_size * seq_length))

    # Drop the last few characters to make only full batches
    xdata = np.array(int_text[: n_batches * batch_size * seq_length])
    ydata = np.array(int_text[1: n_batches * batch_size * seq_length + 1])

    x_batches = np.split(xdata.reshape(batch_size, -1), n_batches, 1)
    y_batches = np.split(ydata.reshape(batch_size, -1), n_batches, 1)

    return np.asarray(list(zip(x_batches, y_batches)))

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 [30]:
# Number of Epochs
num_epochs = 200
# Batch Size
batch_size = 64
# RNN Size
rnn_size = 512
# Sequence Length
seq_length = 32
# Learning Rate
learning_rate = 0.001
# Show stats for every n number of batches
show_every_n_batches = 15

save_dir = './save'

Build the Graph

Build the graph using the neural network we implemented.


In [31]:
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.


In [32]:
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    0/33   train_loss = 8.823
Epoch   0 Batch   15/33   train_loss = 6.543
Epoch   0 Batch   30/33   train_loss = 6.313
Epoch   1 Batch   12/33   train_loss = 6.066
Epoch   1 Batch   27/33   train_loss = 6.219
Epoch   2 Batch    9/33   train_loss = 6.132
Epoch   2 Batch   24/33   train_loss = 6.131
Epoch   3 Batch    6/33   train_loss = 6.013
Epoch   3 Batch   21/33   train_loss = 5.985
Epoch   4 Batch    3/33   train_loss = 6.028
Epoch   4 Batch   18/33   train_loss = 5.905
Epoch   5 Batch    0/33   train_loss = 5.790
Epoch   5 Batch   15/33   train_loss = 5.901
Epoch   5 Batch   30/33   train_loss = 5.809
Epoch   6 Batch   12/33   train_loss = 5.757
Epoch   6 Batch   27/33   train_loss = 5.848
Epoch   7 Batch    9/33   train_loss = 5.757
Epoch   7 Batch   24/33   train_loss = 5.735
Epoch   8 Batch    6/33   train_loss = 5.631
Epoch   8 Batch   21/33   train_loss = 5.589
Epoch   9 Batch    3/33   train_loss = 5.617
Epoch   9 Batch   18/33   train_loss = 5.492
Epoch  10 Batch    0/33   train_loss = 5.329
Epoch  10 Batch   15/33   train_loss = 5.477
Epoch  10 Batch   30/33   train_loss = 5.337
Epoch  11 Batch   12/33   train_loss = 5.275
Epoch  11 Batch   27/33   train_loss = 5.318
Epoch  12 Batch    9/33   train_loss = 5.186
Epoch  12 Batch   24/33   train_loss = 5.183
Epoch  13 Batch    6/33   train_loss = 5.055
Epoch  13 Batch   21/33   train_loss = 4.970
Epoch  14 Batch    3/33   train_loss = 4.957
Epoch  14 Batch   18/33   train_loss = 4.880
Epoch  15 Batch    0/33   train_loss = 4.715
Epoch  15 Batch   15/33   train_loss = 4.905
Epoch  15 Batch   30/33   train_loss = 4.839
Epoch  16 Batch   12/33   train_loss = 4.754
Epoch  16 Batch   27/33   train_loss = 4.832
Epoch  17 Batch    9/33   train_loss = 4.689
Epoch  17 Batch   24/33   train_loss = 4.736
Epoch  18 Batch    6/33   train_loss = 4.613
Epoch  18 Batch   21/33   train_loss = 4.591
Epoch  19 Batch    3/33   train_loss = 4.573
Epoch  19 Batch   18/33   train_loss = 4.509
Epoch  20 Batch    0/33   train_loss = 4.387
Epoch  20 Batch   15/33   train_loss = 4.553
Epoch  20 Batch   30/33   train_loss = 4.549
Epoch  21 Batch   12/33   train_loss = 4.472
Epoch  21 Batch   27/33   train_loss = 4.496
Epoch  22 Batch    9/33   train_loss = 4.410
Epoch  22 Batch   24/33   train_loss = 4.442
Epoch  23 Batch    6/33   train_loss = 4.340
Epoch  23 Batch   21/33   train_loss = 4.321
Epoch  24 Batch    3/33   train_loss = 4.315
Epoch  24 Batch   18/33   train_loss = 4.234
Epoch  25 Batch    0/33   train_loss = 4.126
Epoch  25 Batch   15/33   train_loss = 4.289
Epoch  25 Batch   30/33   train_loss = 4.303
Epoch  26 Batch   12/33   train_loss = 4.210
Epoch  26 Batch   27/33   train_loss = 4.246
Epoch  27 Batch    9/33   train_loss = 4.169
Epoch  27 Batch   24/33   train_loss = 4.217
Epoch  28 Batch    6/33   train_loss = 4.120
Epoch  28 Batch   21/33   train_loss = 4.089
Epoch  29 Batch    3/33   train_loss = 4.079
Epoch  29 Batch   18/33   train_loss = 4.018
Epoch  30 Batch    0/33   train_loss = 3.923
Epoch  30 Batch   15/33   train_loss = 4.073
Epoch  30 Batch   30/33   train_loss = 4.082
Epoch  31 Batch   12/33   train_loss = 4.090
Epoch  31 Batch   27/33   train_loss = 4.044
Epoch  32 Batch    9/33   train_loss = 3.980
Epoch  32 Batch   24/33   train_loss = 4.020
Epoch  33 Batch    6/33   train_loss = 3.954
Epoch  33 Batch   21/33   train_loss = 3.882
Epoch  34 Batch    3/33   train_loss = 3.884
Epoch  34 Batch   18/33   train_loss = 3.842
Epoch  35 Batch    0/33   train_loss = 3.749
Epoch  35 Batch   15/33   train_loss = 3.908
Epoch  35 Batch   30/33   train_loss = 3.939
Epoch  36 Batch   12/33   train_loss = 3.875
Epoch  36 Batch   27/33   train_loss = 3.838
Epoch  37 Batch    9/33   train_loss = 3.784
Epoch  37 Batch   24/33   train_loss = 3.827
Epoch  38 Batch    6/33   train_loss = 3.784
Epoch  38 Batch   21/33   train_loss = 3.734
Epoch  39 Batch    3/33   train_loss = 3.735
Epoch  39 Batch   18/33   train_loss = 3.659
Epoch  40 Batch    0/33   train_loss = 3.560
Epoch  40 Batch   15/33   train_loss = 3.737
Epoch  40 Batch   30/33   train_loss = 3.798
Epoch  41 Batch   12/33   train_loss = 3.693
Epoch  41 Batch   27/33   train_loss = 3.668
Epoch  42 Batch    9/33   train_loss = 3.633
Epoch  42 Batch   24/33   train_loss = 3.643
Epoch  43 Batch    6/33   train_loss = 3.584
Epoch  43 Batch   21/33   train_loss = 3.616
Epoch  44 Batch    3/33   train_loss = 3.515
Epoch  44 Batch   18/33   train_loss = 3.445
Epoch  45 Batch    0/33   train_loss = 3.375
Epoch  45 Batch   15/33   train_loss = 3.535
Epoch  45 Batch   30/33   train_loss = 3.581
Epoch  46 Batch   12/33   train_loss = 3.494
Epoch  46 Batch   27/33   train_loss = 3.413
Epoch  47 Batch    9/33   train_loss = 3.388
Epoch  47 Batch   24/33   train_loss = 3.460
Epoch  48 Batch    6/33   train_loss = 3.402
Epoch  48 Batch   21/33   train_loss = 3.413
Epoch  49 Batch    3/33   train_loss = 3.360
Epoch  49 Batch   18/33   train_loss = 3.346
Epoch  50 Batch    0/33   train_loss = 3.265
Epoch  50 Batch   15/33   train_loss = 3.368
Epoch  50 Batch   30/33   train_loss = 3.403
Epoch  51 Batch   12/33   train_loss = 3.423
Epoch  51 Batch   27/33   train_loss = 3.326
Epoch  52 Batch    9/33   train_loss = 3.263
Epoch  52 Batch   24/33   train_loss = 3.322
Epoch  53 Batch    6/33   train_loss = 3.235
Epoch  53 Batch   21/33   train_loss = 3.270
Epoch  54 Batch    3/33   train_loss = 3.164
Epoch  54 Batch   18/33   train_loss = 3.149
Epoch  55 Batch    0/33   train_loss = 3.057
Epoch  55 Batch   15/33   train_loss = 3.179
Epoch  55 Batch   30/33   train_loss = 3.225
Epoch  56 Batch   12/33   train_loss = 3.157
Epoch  56 Batch   27/33   train_loss = 3.097
Epoch  57 Batch    9/33   train_loss = 2.986
Epoch  57 Batch   24/33   train_loss = 3.077
Epoch  58 Batch    6/33   train_loss = 3.099
Epoch  58 Batch   21/33   train_loss = 3.096
Epoch  59 Batch    3/33   train_loss = 2.988
Epoch  59 Batch   18/33   train_loss = 3.002
Epoch  60 Batch    0/33   train_loss = 2.915
Epoch  60 Batch   15/33   train_loss = 3.019
Epoch  60 Batch   30/33   train_loss = 3.098
Epoch  61 Batch   12/33   train_loss = 2.993
Epoch  61 Batch   27/33   train_loss = 2.944
Epoch  62 Batch    9/33   train_loss = 2.901
Epoch  62 Batch   24/33   train_loss = 2.962
Epoch  63 Batch    6/33   train_loss = 2.973
Epoch  63 Batch   21/33   train_loss = 3.062
Epoch  64 Batch    3/33   train_loss = 2.871
Epoch  64 Batch   18/33   train_loss = 2.928
Epoch  65 Batch    0/33   train_loss = 2.845
Epoch  65 Batch   15/33   train_loss = 2.920
Epoch  65 Batch   30/33   train_loss = 2.961
Epoch  66 Batch   12/33   train_loss = 2.898
Epoch  66 Batch   27/33   train_loss = 2.804
Epoch  67 Batch    9/33   train_loss = 2.707
Epoch  67 Batch   24/33   train_loss = 2.829
Epoch  68 Batch    6/33   train_loss = 2.835
Epoch  68 Batch   21/33   train_loss = 2.823
Epoch  69 Batch    3/33   train_loss = 2.704
Epoch  69 Batch   18/33   train_loss = 2.752
Epoch  70 Batch    0/33   train_loss = 2.698
Epoch  70 Batch   15/33   train_loss = 2.784
Epoch  70 Batch   30/33   train_loss = 2.856
Epoch  71 Batch   12/33   train_loss = 2.699
Epoch  71 Batch   27/33   train_loss = 2.660
Epoch  72 Batch    9/33   train_loss = 2.534
Epoch  72 Batch   24/33   train_loss = 2.689
Epoch  73 Batch    6/33   train_loss = 2.643
Epoch  73 Batch   21/33   train_loss = 2.639
Epoch  74 Batch    3/33   train_loss = 2.542
Epoch  74 Batch   18/33   train_loss = 2.554
Epoch  75 Batch    0/33   train_loss = 2.503
Epoch  75 Batch   15/33   train_loss = 2.556
Epoch  75 Batch   30/33   train_loss = 2.615
Epoch  76 Batch   12/33   train_loss = 2.592
Epoch  76 Batch   27/33   train_loss = 2.501
Epoch  77 Batch    9/33   train_loss = 2.393
Epoch  77 Batch   24/33   train_loss = 2.508
Epoch  78 Batch    6/33   train_loss = 2.531
Epoch  78 Batch   21/33   train_loss = 2.538
Epoch  79 Batch    3/33   train_loss = 2.414
Epoch  79 Batch   18/33   train_loss = 2.473
Epoch  80 Batch    0/33   train_loss = 2.394
Epoch  80 Batch   15/33   train_loss = 2.459
Epoch  80 Batch   30/33   train_loss = 2.635
Epoch  81 Batch   12/33   train_loss = 2.510
Epoch  81 Batch   27/33   train_loss = 2.410
Epoch  82 Batch    9/33   train_loss = 2.372
Epoch  82 Batch   24/33   train_loss = 2.365
Epoch  83 Batch    6/33   train_loss = 2.377
Epoch  83 Batch   21/33   train_loss = 2.393
Epoch  84 Batch    3/33   train_loss = 2.324
Epoch  84 Batch   18/33   train_loss = 2.309
Epoch  85 Batch    0/33   train_loss = 2.221
Epoch  85 Batch   15/33   train_loss = 2.310
Epoch  85 Batch   30/33   train_loss = 2.373
Epoch  86 Batch   12/33   train_loss = 2.303
Epoch  86 Batch   27/33   train_loss = 2.219
Epoch  87 Batch    9/33   train_loss = 2.115
Epoch  87 Batch   24/33   train_loss = 2.228
Epoch  88 Batch    6/33   train_loss = 2.224
Epoch  88 Batch   21/33   train_loss = 2.250
Epoch  89 Batch    3/33   train_loss = 2.093
Epoch  89 Batch   18/33   train_loss = 2.153
Epoch  90 Batch    0/33   train_loss = 2.109
Epoch  90 Batch   15/33   train_loss = 2.115
Epoch  90 Batch   30/33   train_loss = 2.203
Epoch  91 Batch   12/33   train_loss = 2.154
Epoch  91 Batch   27/33   train_loss = 2.071
Epoch  92 Batch    9/33   train_loss = 1.979
Epoch  92 Batch   24/33   train_loss = 2.094
Epoch  93 Batch    6/33   train_loss = 2.111
Epoch  93 Batch   21/33   train_loss = 2.096
Epoch  94 Batch    3/33   train_loss = 2.024
Epoch  94 Batch   18/33   train_loss = 2.023
Epoch  95 Batch    0/33   train_loss = 1.956
Epoch  95 Batch   15/33   train_loss = 2.029
Epoch  95 Batch   30/33   train_loss = 2.097
Epoch  96 Batch   12/33   train_loss = 2.095
Epoch  96 Batch   27/33   train_loss = 1.996
Epoch  97 Batch    9/33   train_loss = 1.862
Epoch  97 Batch   24/33   train_loss = 1.932
Epoch  98 Batch    6/33   train_loss = 1.976
Epoch  98 Batch   21/33   train_loss = 2.029
Epoch  99 Batch    3/33   train_loss = 1.928
Epoch  99 Batch   18/33   train_loss = 1.961
Epoch 100 Batch    0/33   train_loss = 1.840
Epoch 100 Batch   15/33   train_loss = 1.900
Epoch 100 Batch   30/33   train_loss = 2.001
Epoch 101 Batch   12/33   train_loss = 1.914
Epoch 101 Batch   27/33   train_loss = 1.841
Epoch 102 Batch    9/33   train_loss = 1.757
Epoch 102 Batch   24/33   train_loss = 1.839
Epoch 103 Batch    6/33   train_loss = 1.901
Epoch 103 Batch   21/33   train_loss = 1.908
Epoch 104 Batch    3/33   train_loss = 1.804
Epoch 104 Batch   18/33   train_loss = 1.806
Epoch 105 Batch    0/33   train_loss = 1.740
Epoch 105 Batch   15/33   train_loss = 1.795
Epoch 105 Batch   30/33   train_loss = 1.925
Epoch 106 Batch   12/33   train_loss = 1.880
Epoch 106 Batch   27/33   train_loss = 1.808
Epoch 107 Batch    9/33   train_loss = 1.634
Epoch 107 Batch   24/33   train_loss = 1.804
Epoch 108 Batch    6/33   train_loss = 1.802
Epoch 108 Batch   21/33   train_loss = 1.816
Epoch 109 Batch    3/33   train_loss = 1.715
Epoch 109 Batch   18/33   train_loss = 1.729
Epoch 110 Batch    0/33   train_loss = 1.713
Epoch 110 Batch   15/33   train_loss = 1.723
Epoch 110 Batch   30/33   train_loss = 1.796
Epoch 111 Batch   12/33   train_loss = 1.859
Epoch 111 Batch   27/33   train_loss = 1.796
Epoch 112 Batch    9/33   train_loss = 1.568
Epoch 112 Batch   24/33   train_loss = 1.772
Epoch 113 Batch    6/33   train_loss = 1.799
Epoch 113 Batch   21/33   train_loss = 1.766
Epoch 114 Batch    3/33   train_loss = 1.653
Epoch 114 Batch   18/33   train_loss = 1.618
Epoch 115 Batch    0/33   train_loss = 1.615
Epoch 115 Batch   15/33   train_loss = 1.607
Epoch 115 Batch   30/33   train_loss = 1.690
Epoch 116 Batch   12/33   train_loss = 1.656
Epoch 116 Batch   27/33   train_loss = 1.610
Epoch 117 Batch    9/33   train_loss = 1.491
Epoch 117 Batch   24/33   train_loss = 1.545
Epoch 118 Batch    6/33   train_loss = 1.635
Epoch 118 Batch   21/33   train_loss = 1.630
Epoch 119 Batch    3/33   train_loss = 1.513
Epoch 119 Batch   18/33   train_loss = 1.563
Epoch 120 Batch    0/33   train_loss = 1.505
Epoch 120 Batch   15/33   train_loss = 1.502
Epoch 120 Batch   30/33   train_loss = 1.611
Epoch 121 Batch   12/33   train_loss = 1.582
Epoch 121 Batch   27/33   train_loss = 1.500
Epoch 122 Batch    9/33   train_loss = 1.406
Epoch 122 Batch   24/33   train_loss = 1.467
Epoch 123 Batch    6/33   train_loss = 1.561
Epoch 123 Batch   21/33   train_loss = 1.587
Epoch 124 Batch    3/33   train_loss = 1.431
Epoch 124 Batch   18/33   train_loss = 1.404
Epoch 125 Batch    0/33   train_loss = 1.413
Epoch 125 Batch   15/33   train_loss = 1.440
Epoch 125 Batch   30/33   train_loss = 1.502
Epoch 126 Batch   12/33   train_loss = 1.498
Epoch 126 Batch   27/33   train_loss = 1.389
Epoch 127 Batch    9/33   train_loss = 1.376
Epoch 127 Batch   24/33   train_loss = 1.428
Epoch 128 Batch    6/33   train_loss = 1.474
Epoch 128 Batch   21/33   train_loss = 1.480
Epoch 129 Batch    3/33   train_loss = 1.380
Epoch 129 Batch   18/33   train_loss = 1.362
Epoch 130 Batch    0/33   train_loss = 1.335
Epoch 130 Batch   15/33   train_loss = 1.372
Epoch 130 Batch   30/33   train_loss = 1.441
Epoch 131 Batch   12/33   train_loss = 1.455
Epoch 131 Batch   27/33   train_loss = 1.339
Epoch 132 Batch    9/33   train_loss = 1.267
Epoch 132 Batch   24/33   train_loss = 1.281
Epoch 133 Batch    6/33   train_loss = 1.373
Epoch 133 Batch   21/33   train_loss = 1.437
Epoch 134 Batch    3/33   train_loss = 1.283
Epoch 134 Batch   18/33   train_loss = 1.338
Epoch 135 Batch    0/33   train_loss = 1.296
Epoch 135 Batch   15/33   train_loss = 1.290
Epoch 135 Batch   30/33   train_loss = 1.355
Epoch 136 Batch   12/33   train_loss = 1.407
Epoch 136 Batch   27/33   train_loss = 1.260
Epoch 137 Batch    9/33   train_loss = 1.215
Epoch 137 Batch   24/33   train_loss = 1.261
Epoch 138 Batch    6/33   train_loss = 1.283
Epoch 138 Batch   21/33   train_loss = 1.358
Epoch 139 Batch    3/33   train_loss = 1.184
Epoch 139 Batch   18/33   train_loss = 1.232
Epoch 140 Batch    0/33   train_loss = 1.211
Epoch 140 Batch   15/33   train_loss = 1.250
Epoch 140 Batch   30/33   train_loss = 1.304
Epoch 141 Batch   12/33   train_loss = 1.273
Epoch 141 Batch   27/33   train_loss = 1.190
Epoch 142 Batch    9/33   train_loss = 1.095
Epoch 142 Batch   24/33   train_loss = 1.151
Epoch 143 Batch    6/33   train_loss = 1.233
Epoch 143 Batch   21/33   train_loss = 1.294
Epoch 144 Batch    3/33   train_loss = 1.162
Epoch 144 Batch   18/33   train_loss = 1.185
Epoch 145 Batch    0/33   train_loss = 1.081
Epoch 145 Batch   15/33   train_loss = 1.125
Epoch 145 Batch   30/33   train_loss = 1.212
Epoch 146 Batch   12/33   train_loss = 1.280
Epoch 146 Batch   27/33   train_loss = 1.168
Epoch 147 Batch    9/33   train_loss = 1.068
Epoch 147 Batch   24/33   train_loss = 1.105
Epoch 148 Batch    6/33   train_loss = 1.159
Epoch 148 Batch   21/33   train_loss = 1.206
Epoch 149 Batch    3/33   train_loss = 1.109
Epoch 149 Batch   18/33   train_loss = 1.110
Epoch 150 Batch    0/33   train_loss = 1.082
Epoch 150 Batch   15/33   train_loss = 1.129
Epoch 150 Batch   30/33   train_loss = 1.177
Epoch 151 Batch   12/33   train_loss = 1.191
Epoch 151 Batch   27/33   train_loss = 1.073
Epoch 152 Batch    9/33   train_loss = 1.025
Epoch 152 Batch   24/33   train_loss = 1.117
Epoch 153 Batch    6/33   train_loss = 1.143
Epoch 153 Batch   21/33   train_loss = 1.190
Epoch 154 Batch    3/33   train_loss = 1.154
Epoch 154 Batch   18/33   train_loss = 1.088
Epoch 155 Batch    0/33   train_loss = 1.046
Epoch 155 Batch   15/33   train_loss = 1.125
Epoch 155 Batch   30/33   train_loss = 1.165
Epoch 156 Batch   12/33   train_loss = 1.143
Epoch 156 Batch   27/33   train_loss = 1.041
Epoch 157 Batch    9/33   train_loss = 0.961
Epoch 157 Batch   24/33   train_loss = 1.005
Epoch 158 Batch    6/33   train_loss = 1.067
Epoch 158 Batch   21/33   train_loss = 1.069
Epoch 159 Batch    3/33   train_loss = 0.950
Epoch 159 Batch   18/33   train_loss = 1.000
Epoch 160 Batch    0/33   train_loss = 0.945
Epoch 160 Batch   15/33   train_loss = 0.987
Epoch 160 Batch   30/33   train_loss = 1.026
Epoch 161 Batch   12/33   train_loss = 1.041
Epoch 161 Batch   27/33   train_loss = 0.923
Epoch 162 Batch    9/33   train_loss = 0.881
Epoch 162 Batch   24/33   train_loss = 0.920
Epoch 163 Batch    6/33   train_loss = 0.986
Epoch 163 Batch   21/33   train_loss = 1.013
Epoch 164 Batch    3/33   train_loss = 0.901
Epoch 164 Batch   18/33   train_loss = 0.939
Epoch 165 Batch    0/33   train_loss = 0.890
Epoch 165 Batch   15/33   train_loss = 0.945
Epoch 165 Batch   30/33   train_loss = 0.973
Epoch 166 Batch   12/33   train_loss = 1.010
Epoch 166 Batch   27/33   train_loss = 0.889
Epoch 167 Batch    9/33   train_loss = 0.825
Epoch 167 Batch   24/33   train_loss = 0.876
Epoch 168 Batch    6/33   train_loss = 0.916
Epoch 168 Batch   21/33   train_loss = 0.928
Epoch 169 Batch    3/33   train_loss = 0.864
Epoch 169 Batch   18/33   train_loss = 0.896
Epoch 170 Batch    0/33   train_loss = 0.825
Epoch 170 Batch   15/33   train_loss = 0.880
Epoch 170 Batch   30/33   train_loss = 0.914
Epoch 171 Batch   12/33   train_loss = 0.929
Epoch 171 Batch   27/33   train_loss = 0.864
Epoch 172 Batch    9/33   train_loss = 0.806
Epoch 172 Batch   24/33   train_loss = 0.863
Epoch 173 Batch    6/33   train_loss = 0.916
Epoch 173 Batch   21/33   train_loss = 0.940
Epoch 174 Batch    3/33   train_loss = 0.838
Epoch 174 Batch   18/33   train_loss = 0.860
Epoch 175 Batch    0/33   train_loss = 0.828
Epoch 175 Batch   15/33   train_loss = 0.860
Epoch 175 Batch   30/33   train_loss = 0.887
Epoch 176 Batch   12/33   train_loss = 0.935
Epoch 176 Batch   27/33   train_loss = 0.849
Epoch 177 Batch    9/33   train_loss = 0.785
Epoch 177 Batch   24/33   train_loss = 0.816
Epoch 178 Batch    6/33   train_loss = 0.861
Epoch 178 Batch   21/33   train_loss = 0.881
Epoch 179 Batch    3/33   train_loss = 0.810
Epoch 179 Batch   18/33   train_loss = 0.861
Epoch 180 Batch    0/33   train_loss = 0.767
Epoch 180 Batch   15/33   train_loss = 0.840
Epoch 180 Batch   30/33   train_loss = 0.872
Epoch 181 Batch   12/33   train_loss = 0.892
Epoch 181 Batch   27/33   train_loss = 0.795
Epoch 182 Batch    9/33   train_loss = 0.703
Epoch 182 Batch   24/33   train_loss = 0.772
Epoch 183 Batch    6/33   train_loss = 0.830
Epoch 183 Batch   21/33   train_loss = 0.826
Epoch 184 Batch    3/33   train_loss = 0.723
Epoch 184 Batch   18/33   train_loss = 0.790
Epoch 185 Batch    0/33   train_loss = 0.721
Epoch 185 Batch   15/33   train_loss = 0.785
Epoch 185 Batch   30/33   train_loss = 0.781
Epoch 186 Batch   12/33   train_loss = 0.851
Epoch 186 Batch   27/33   train_loss = 0.744
Epoch 187 Batch    9/33   train_loss = 0.699
Epoch 187 Batch   24/33   train_loss = 0.705
Epoch 188 Batch    6/33   train_loss = 0.775
Epoch 188 Batch   21/33   train_loss = 0.765
Epoch 189 Batch    3/33   train_loss = 0.694
Epoch 189 Batch   18/33   train_loss = 0.746
Epoch 190 Batch    0/33   train_loss = 0.710
Epoch 190 Batch   15/33   train_loss = 0.705
Epoch 190 Batch   30/33   train_loss = 0.753
Epoch 191 Batch   12/33   train_loss = 0.786
Epoch 191 Batch   27/33   train_loss = 0.680
Epoch 192 Batch    9/33   train_loss = 0.657
Epoch 192 Batch   24/33   train_loss = 0.647
Epoch 193 Batch    6/33   train_loss = 0.752
Epoch 193 Batch   21/33   train_loss = 0.712
Epoch 194 Batch    3/33   train_loss = 0.639
Epoch 194 Batch   18/33   train_loss = 0.701
Epoch 195 Batch    0/33   train_loss = 0.623
Epoch 195 Batch   15/33   train_loss = 0.670
Epoch 195 Batch   30/33   train_loss = 0.724
Epoch 196 Batch   12/33   train_loss = 0.740
Epoch 196 Batch   27/33   train_loss = 0.669
Epoch 197 Batch    9/33   train_loss = 0.629
Epoch 197 Batch   24/33   train_loss = 0.635
Epoch 198 Batch    6/33   train_loss = 0.690
Epoch 198 Batch   21/33   train_loss = 0.691
Epoch 199 Batch    3/33   train_loss = 0.623
Epoch 199 Batch   18/33   train_loss = 0.661
Model Trained and Saved

Save Parameters

Save seq_length and save_dir for generating a new TV script.


In [33]:
# Save parameters for checkpoint
helper.save_params((seq_length, save_dir))

Checkpoint


In [34]:
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().

Return the tensors in the following tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)


In [35]:
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)
    """
    # 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


tests.test_get_tensors(get_tensors)


Tests Passed

Choose Word

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


In [36]:
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
    """
    # Implement Function
    #int_words = [int_to_vocab[word] for word in text]
    #train_words = [word for word in int_to_vocab if np.any(probabilities < random.random())]
    
    index = np.argmax(probabilities) 
    train_words = int_to_vocab[index]
    
    return train_words


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 we want to generate.


In [37]:
gen_length = 200
# homer_simpson, moe_szyslak, or Barney_Gumble
prime_word = 'moe_szyslak'

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: uh...(ad lib singing) my bar could be british / instead of arm-pittish / all my problem.
moe_szyslak: noooooooooo!
homer_simpson: what's the matter, moe?
moe_szyslak: okay, he's opening my car?


homer's_brain: eh, no matter how much kid could my money. i'm too bad.
lenny_leonard: no way how like it.(uneasy) oh, moe, she asked for your love channel before i love some milk?
moe_szyslak: uh, how like you were looking for a kid young thing with our head, he simpson.
homer_simpson: no, i don't have a new kids that might all bad into the house.
bart_simpson: well, what-- i've been replaced by this one out of a good one.(pretends to be second book) no. i was gonna break as this place(chuckles)...
moe_szyslak: ah, uh, check. but we gotta find a new guy in the hell channel he made it like.
apu_nahasapeemapetilon: the wants to understanding by

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, we'll have to use a smaller vocabulary or get more data. This was a subset of another dataset. We can use it all!