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.
%%html
<style>
@import url('https://fonts.googleapis.com/css?family=Orbitron|Roboto');
body {background-color: aliceblue;}
a {color: darksteelblue; font-family: Roboto;}
h1, h2 {color: #338DD4; font-family: Orbitron; text-shadow: 4px 4px 4px #ccc;}
h3, h4 {color: steelblue; font-family: Roboto; text-shadow: 4px 4px 4px #ccc;}
span {text-shadow: 4px 4px 4px #ccc;}
div.output_prompt, div.output_area pre {color: slategray;}
div.input_prompt, div.output_subarea {color: darksteelblue;}
div.output_stderr pre {background-color: aliceblue;}
div.output_stderr {background-color: slategrey;}
</style>
<script>
code_show = true;
function code_display() {
if (code_show) {
$('div.input').each(function(id) {
if (id == 0 || $(this).html().indexOf('hide_code') > -1) {$(this).hide();}
});
$('div.output_prompt').css('opacity', 0);
} else {
$('div.input').each(function(id) {$(this).show();});
$('div.output_prompt').css('opacity', 1);
};
code_show = !code_show;
}
$(document).ready(code_display);
</script>
<form action="javascript: code_display()">
<input style="color: #338DD4; background: aliceblue; text-shadow: 4px 4px 4px #ccc;" \
type="submit" value="Click to display or hide code cells">
</form>
hide_code=''
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import numpy as np
import os
import pickle
import tensorflow as tf
from tensorflow.contrib import rnn
from tensorflow.contrib import seq2seq
from distutils.version import LooseVersion
import warnings
hide_code
# https://github.com/udacity/deep-learning/blob/master/tv-script-generation/helper.py
def load_data(path):
"""Load Dataset from File"""
with open(os.path.join(path), "r") as f:
data = f.read()
return data
def preprocess_and_save_data(dataset_path, token_lookup, create_lookup_tables):
"""Preprocess Text Data"""
text = load_data(dataset_path)
# Ignore notice, since we don't use it for analysing the data
text = text[81:]
token_dict = token_lookup()
for key, token in token_dict.items():
text = text.replace(key, ' {} '.format(token))
text = text.lower()
text = text.split()
vocab_to_int, int_to_vocab = create_lookup_tables(text)
int_text = [vocab_to_int[word] for word in text]
pickle.dump((int_text, vocab_to_int, int_to_vocab, token_dict), open('preprocess.p', 'wb'))
def load_preprocess():
"""Load the Preprocessed Training data and return them in batches of <batch_size> or less"""
return pickle.load(open('preprocess.p', mode='rb'))
def save_params(params):
"""Save parameters to file"""
pickle.dump(params, open('params.p', 'wb'))
def load_params():
"""Load parameters from file"""
return pickle.load(open('params.p', mode='rb'))
hide_code
# https://github.com/udacity/deep-learning/blob/master/tv-script-generation/problem_unittests.py
def _print_success_message():
print('Tests Passed')
def test_create_lookup_tables(create_lookup_tables):
with tf.Graph().as_default():
test_text = '''
Moe_Szyslak Moe's Tavern Where the elite meet to drink
Bart_Simpson Eh yeah hello is Mike there Last name Rotch
Moe_Szyslak Hold on I'll check Mike Rotch Mike Rotch Hey has anybody seen Mike Rotch lately
Moe_Szyslak 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 Whats 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'''
test_text = test_text.lower()
test_text = test_text.split()
vocab_to_int, int_to_vocab = create_lookup_tables(test_text)
# Check types
assert isinstance(vocab_to_int, dict),\
'vocab_to_int is not a dictionary.'
assert isinstance(int_to_vocab, dict),\
'int_to_vocab is not a dictionary.'
# Compare lengths of dicts
assert len(vocab_to_int) == len(int_to_vocab),\
'Length of vocab_to_int and int_to_vocab don\'t match. ' \
'vocab_to_int is length {}. int_to_vocab is length {}'.format(len(vocab_to_int), len(int_to_vocab))
# Make sure the dicts have the same words
vocab_to_int_word_set = set(vocab_to_int.keys())
int_to_vocab_word_set = set(int_to_vocab.values())
assert not (vocab_to_int_word_set - int_to_vocab_word_set),\
'vocab_to_int and int_to_vocab don\'t have the same words.' \
'{} found in vocab_to_int, but not in int_to_vocab'\
.format(vocab_to_int_word_set - int_to_vocab_word_set)
assert not (int_to_vocab_word_set - vocab_to_int_word_set),\
'vocab_to_int and int_to_vocab don\'t have the same words.' \
'{} found in int_to_vocab, but not in vocab_to_int'\
.format(int_to_vocab_word_set - vocab_to_int_word_set)
# Make sure the dicts have the same word ids
vocab_to_int_word_id_set = set(vocab_to_int.values())
int_to_vocab_word_id_set = set(int_to_vocab.keys())
assert not (vocab_to_int_word_id_set - int_to_vocab_word_id_set),\
'vocab_to_int and int_to_vocab don\'t contain the same word ids.' \
'{} found in vocab_to_int, but not in int_to_vocab'\
.format(vocab_to_int_word_id_set - int_to_vocab_word_id_set)
assert not (int_to_vocab_word_id_set - vocab_to_int_word_id_set),\
'vocab_to_int and int_to_vocab don\'t contain the same word ids.' \
'{} found in int_to_vocab, but not in vocab_to_int'\
.format(int_to_vocab_word_id_set - vocab_to_int_word_id_set)
# Make sure the dicts make the same lookup
missmatches = [(word, id, id, int_to_vocab[id]) \
for word, id in vocab_to_int.items() if int_to_vocab[id] != word]
assert not missmatches,\
'Found {} missmatche(s). First missmatch: vocab_to_int[{}] = {} and int_to_vocab[{}] = {}'\
.format(len(missmatches),*missmatches[0])
assert len(vocab_to_int) > len(set(test_text))/2,\
'The length of vocab seems too small. Found a length of {}'.format(len(vocab_to_int))
_print_success_message()
def test_get_batches(get_batches):
with tf.Graph().as_default():
test_batch_size = 128
test_seq_length = 5
test_int_text = list(range(1000*test_seq_length))
batches = get_batches(test_int_text, test_batch_size, test_seq_length)
# Check type
assert isinstance(batches, np.ndarray),\
'Batches is not a Numpy array'
# Check shape
assert batches.shape == (7, 2, 128, 5),\
'Batches returned wrong shape. Found {}'.format(batches.shape)
for x in range(batches.shape[2]):
assert np.array_equal(batches[0,0,x], np.array(range(x * 35, x * 35 + batches.shape[3]))),\
'Batches returned wrong contents. For example, input sequence {} in the first batch was {}'\
.format(x, batches[0,0,x])
assert np.array_equal(batches[0,1,x], np.array(range(x * 35 + 1, x * 35 + 1 + batches.shape[3]))),\
'Batches returned wrong contents. For example, target sequence {} in the first batch was {}'\
.format(x, batches[0,1,x])
last_seq_target = (test_batch_size-1) * 35 + 31
last_seq = np.array(range(last_seq_target, last_seq_target+ batches.shape[3]))
last_seq[-1] = batches[0,0,0,0]
assert np.array_equal(batches[-1,1,-1], last_seq),\
'The last target of the last batch should be the first input of the first batch. \
Found {} but expected {}'.format(batches[-1,1,-1], last_seq)
_print_success_message()
def test_tokenize(token_lookup):
with tf.Graph().as_default():
symbols = set(['.', ',', '"', ';', '!', '?', '(', ')', '--', '\n'])
token_dict = token_lookup()
# Check type
assert isinstance(token_dict, dict), \
'Returned type is {}.'.format(type(token_dict))
# Check symbols
missing_symbols = symbols - set(token_dict.keys())
unknown_symbols = set(token_dict.keys()) - symbols
assert not missing_symbols, \
'Missing symbols: {}'.format(missing_symbols)
assert not unknown_symbols, \
'Unknown symbols: {}'.format(unknown_symbols)
# Check values type
bad_value_type = [type(val) for val in token_dict.values() if not isinstance(val, str)]
assert not bad_value_type,\
'Found token as {} type.'.format(bad_value_type[0])
# Check for spaces
key_has_spaces = [k for k in token_dict.keys() if ' ' in k]
val_has_spaces = [val for val in token_dict.values() if ' ' in val]
assert not key_has_spaces,\
'The key "{}" includes spaces. Remove spaces from keys and values'.format(key_has_spaces[0])
assert not val_has_spaces,\
'The value "{}" includes spaces. Remove spaces from keys and values'.format(val_has_spaces[0])
# Check for symbols in values
symbol_val = ()
for symbol in symbols:
for val in token_dict.values():
if symbol in val:
symbol_val = (symbol, val)
assert not symbol_val,\
'Don\'t use a symbol that will be replaced in your tokens. Found the symbol {} in value {}'\
.format(*symbol_val)
_print_success_message()
def test_get_inputs(get_inputs):
with tf.Graph().as_default():
input_data, targets, lr = get_inputs()
# Check type
assert input_data.op.type == 'Placeholder',\
'Input not a Placeholder.'
assert targets.op.type == 'Placeholder',\
'Targets not a Placeholder.'
assert lr.op.type == 'Placeholder',\
'Learning Rate not a Placeholder.'
# Check name
assert input_data.name == 'input:0',\
'Input has bad name. Found name {}'.format(input_data.name)
# Check rank
input_rank = 0 if input_data.get_shape() == None else len(input_data.get_shape())
targets_rank = 0 if targets.get_shape() == None else len(targets.get_shape())
lr_rank = 0 if lr.get_shape() == None else len(lr.get_shape())
assert input_rank == 2,\
'Input has wrong rank. Rank {} found.'.format(input_rank)
assert targets_rank == 2,\
'Targets has wrong rank. Rank {} found.'.format(targets_rank)
assert lr_rank == 0,\
'Learning Rate has wrong rank. Rank {} found'.format(lr_rank)
_print_success_message()
def test_get_init_cell(get_init_cell):
with tf.Graph().as_default():
test_batch_size_ph = tf.placeholder(tf.int32, [])
test_rnn_size = 256
cell, init_state = get_init_cell(test_batch_size_ph, test_rnn_size)
# Check type
assert isinstance(cell, tf.contrib.rnn.MultiRNNCell),\
'Cell is wrong type. Found {} type'.format(type(cell))
# Check for name attribute
assert hasattr(init_state, 'name'),\
'Initial state doesn\'t have the "name" attribute. Try using `tf.identity` to set the name.'
# Check name
assert init_state.name == 'initial_state:0',\
'Initial state doesn\'t have the correct name. Found the name {}'.format(init_state.name)
_print_success_message()
def test_get_embed(get_embed):
with tf.Graph().as_default():
embed_shape = [50, 5, 256]
test_input_data = tf.placeholder(tf.int32, embed_shape[:2])
test_vocab_size = 27
test_embed_dim = embed_shape[2]
embed = get_embed(test_input_data, test_vocab_size, test_embed_dim)
# Check shape
assert embed.shape == embed_shape,\
'Wrong shape. Found shape {}'.format(embed.shape)
_print_success_message()
def test_build_rnn(build_rnn):
with tf.Graph().as_default():
test_rnn_size = 256
test_rnn_layer_size = 2
test_cell = rnn.MultiRNNCell([rnn.BasicLSTMCell(test_rnn_size) for _ in range(test_rnn_layer_size)])
test_inputs = tf.placeholder(tf.float32, [None, None, test_rnn_size])
outputs, final_state = build_rnn(test_cell, test_inputs)
# Check name
assert hasattr(final_state, 'name'),\
'Final state doesn\'t have the "name" attribute. Try using `tf.identity` to set the name.'
assert final_state.name == 'final_state:0',\
'Final state doesn\'t have the correct name. Found the name {}'.format(final_state.name)
# Check shape
assert outputs.get_shape().as_list() == [None, None, test_rnn_size],\
'Outputs has wrong shape. Found shape {}'.format(outputs.get_shape())
assert final_state.get_shape().as_list() == [test_rnn_layer_size, 2, None, test_rnn_size],\
'Final state wrong shape. Found shape {}'.format(final_state.get_shape())
_print_success_message()
def test_build_nn(build_nn):
with tf.Graph().as_default():
test_input_data_shape = [128, 5]
test_input_data = tf.placeholder(tf.int32, test_input_data_shape)
test_rnn_size = 256
test_embed_dim = 300
test_rnn_layer_size = 2
test_vocab_size = 27
test_cell = rnn.MultiRNNCell([rnn.BasicLSTMCell(test_rnn_size) for _ in range(test_rnn_layer_size)])
logits, final_state = build_nn(test_cell, test_rnn_size, test_input_data, test_vocab_size, test_embed_dim)
# Check name
assert hasattr(final_state, 'name'), \
'Final state doesn\'t have the "name" attribute. Are you using build_rnn?'
assert final_state.name == 'final_state:0', \
'Final state doesn\'t have the correct name. Found the name {}. Are you using build_rnn?'\
.format(final_state.name)
# Check Shape
assert logits.get_shape().as_list() == test_input_data_shape + [test_vocab_size], \
'Outputs has wrong shape. Found shape {}'.format(logits.get_shape())
assert final_state.get_shape().as_list() == [test_rnn_layer_size, 2, None, test_rnn_size], \
'Final state wrong shape. Found shape {}'.format(final_state.get_shape())
_print_success_message()
def test_get_tensors(get_tensors):
test_graph = tf.Graph()
with test_graph.as_default():
test_input = tf.placeholder(tf.int32, name='input')
test_initial_state = tf.placeholder(tf.int32, name='initial_state')
test_final_state = tf.placeholder(tf.int32, name='final_state')
test_probs = tf.placeholder(tf.float32, name='probs')
input_text, initial_state, final_state, probs = get_tensors(test_graph)
# Check correct tensor
assert input_text == test_input,\
'Test input is wrong tensor'
assert initial_state == test_initial_state, \
'Initial state is wrong tensor'
assert final_state == test_final_state, \
'Final state is wrong tensor'
assert probs == test_probs, \
'Probabilities is wrong tensor'
_print_success_message()
def test_pick_word(pick_word):
with tf.Graph().as_default():
test_probabilities = np.array([0.1, 0.8, 0.05, 0.05])
test_int_to_vocab = {word_i: word for word_i, word in enumerate(['this', 'is', 'a', 'test'])}
pred_word = pick_word(test_probabilities, test_int_to_vocab)
# Check type
assert isinstance(pred_word, str),\
'Predicted word is wrong type. Found {} type.'.format(type(pred_word))
# Check word is from vocab
assert pred_word in test_int_to_vocab.values(),\
'Predicted word not found in int_to_vocab.'
_print_success_message()
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..
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
data_dir = 'data/simpsons/moes_tavern_lines.txt'
text = load_data(data_dir)
# Ignore notice, since we don't use it for analysing the data
text = text[81:]
Play around with view_sentence_range to view different parts of the data.
hide_code
view_sentence_range = (0, 15)
"""DON'T MODIFY ANYTHING IN THIS CELL"""
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]]))
The first thing to do to any dataset is preprocessing. Implement the following preprocessing functions below:
To create a word embedding, you first need to transform the words to ids. In this function, create two dictionaries:
vocab_to_intint_to_vocabReturn these dictionaries in the following tuple (vocab_to_int, int_to_vocab)
hide_code
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
vocabulary = set(text)
int_to_vocab = dict(enumerate(vocabulary))
vocab_to_int = dict((v,k) for k,v in int_to_vocab.items())
return vocab_to_int, int_to_vocab
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_create_lookup_tables(create_lookup_tables)
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:
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||".
hide_code
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
punctuation_dict = {'.': '$$dot$$', ',': '$$comma$$', '"': '$$quot_mark$$',
';': '$$semicolon$$', '!': '$$excl_mark$$', '?': '$$question$$',
'(': '$$l_parentheses$$', ')': '$$r_parentheses$$',
'--': '$$dash$$', '\n': '$$return$$'}
return punctuation_dict
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_tokenize(token_lookup)
Running the code cell below will preprocess all the data and save it to file.
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
# Preprocess Training, Validation, and Testing Data
preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)
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.
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
int_text, vocab_to_int, int_to_vocab, token_dict = load_preprocess()
hide_code
print('Example of int_text \n', int_text[:100], '\n')
l = [0,1,2,3,4]
print('Example of vocab_to_int \n', {k:v for k,v in vocab_to_int.items() if v in [0,1,2,3,4]}, '\n')
print('Example of int_to_vocab \n', {k:v for k,v in int_to_vocab.items() if k in [0,1,2,3,4]}, '\n')
print('Example of token_dict \n', token_dict)
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
# 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()))
Implement the get_inputs() function to create TF Placeholders for the Neural Network. It should create the following placeholders:
name parameter.Return the placeholders in the following tuple (Input, Targets, LearningRate)
hide_code
def get_inputs():
"""
Create TF Placeholders for input, targets, and learning rate.
:return: Tuple (input, targets, learning rate)
"""
# TODO: Implement Function
inputs = tf.placeholder(tf.int32, shape=[None, None], name='input') # rank 2
targets = tf.placeholder(tf.int32, shape=[None, None], name='targets') # rank 2
learning_rate = tf.placeholder(tf.float32, shape=[], name='learning_rate') # rank 0
return inputs, targets, learning_rate
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_get_inputs(get_inputs)
Stack one or more BasicLSTMCells in a MultiRNNCell.
rnn_sizezero_state() functiontf.identity()Return the cell and initial state in the following tuple (Cell, InitialState)
hide_code
def get_init_cell(batch_size, rnn_size, keep_prob=0.8, num_layers=2):
"""
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
def create_lstm_cell(state_size):
lstm_layer = tf.contrib.rnn.core_rnn_cell.BasicLSTMCell(state_size)
return tf.contrib.rnn.core_rnn_cell.DropoutWrapper(lstm_layer,
input_keep_prob=1.0,
output_keep_prob=keep_prob)
rnn_cell = tf.contrib.rnn.\
core_rnn_cell.MultiRNNCell([create_lstm_cell(rnn_size) for _ in range(num_layers)])
initial_state = rnn_cell.zero_state(batch_size, tf.float32)
initial_state = tf.identity(initial_state, name='initial_state')
return rnn_cell, initial_state
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_get_init_cell(get_init_cell)
hide_code
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
word_embeddings = tf.Variable(tf.random_uniform((vocab_size, embed_dim), -1, 1))
embedded_word_ids = tf.nn.embedding_lookup(word_embeddings, input_data)
return embedded_word_ids
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_get_embed(get_embed)
You created a RNN Cell in the get_init_cell() function. Time to use the cell to create a RNN.
tf.nn.dynamic_rnn()tf.identity()Return the outputs and final_state state in the following tuple (Outputs, FinalState)
hide_code
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"""
test_build_rnn(build_rnn)
Apply the functions you implemented above to:
input_data using your get_embed(input_data, vocab_size, embed_dim) function.cell and your build_rnn(cell, inputs) function.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)
hide_code
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
emb_outputs = get_embed(input_data, vocab_size, embed_dim)
rnn_outputs, final_state = build_rnn(cell, emb_outputs)
# activation_fn=None == linear activation
logits = tf.contrib.layers.fully_connected(inputs=rnn_outputs,
num_outputs=vocab_size,
activation_fn = None,
weights_initializer = tf.truncated_normal_initializer(stddev=0.1),
biases_initializer = tf.zeros_initializer())
return logits, final_state
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_build_nn(build_nn)
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:
[batch size, sequence length][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, 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.
hide_code
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
slice_size = batch_size * seq_length
n_batches = int(len(int_text) / slice_size)
# targets == features, shifted one character over
X = int_text[: n_batches*slice_size]
y = X[1:] + [X[0]]
X = np.array(X).reshape(batch_size, -1)
y = np.array(y).reshape(batch_size, -1)
X = np.split(X, n_batches, 1)
y = np.split(y, n_batches, 1)
return np.array(list(zip(X, y)))
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_get_batches(get_batches)
Tune the following parameters:
num_epochs to the number of epochs.batch_size to the batch size.rnn_size to the size of the RNNs.embed_dim to the size of the embedding.seq_length to the length of sequence.learning_rate to the learning rate.show_every_n_batches to the number of batches the neural network should print progress.hide_code
num_epochs = 800
batch_size = 128
rnn_size = 256
embed_dim = 128
seq_length = 32
learning_rate = 0.001
# Show stats for every n number of batches
show_every_n_batches = 128
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
save_dir = './save'
Build the graph using the neural network you implemented.
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
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)
hide_code
"""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')
Save seq_length and save_dir for generating a new TV script.
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
# Save parameters for checkpoint
save_params((seq_length, save_dir))
hide_code
"""DON'T MODIFY ANYTHING IN THIS CELL"""
_, vocab_to_int, int_to_vocab, token_dict = load_preprocess()
seq_length, load_dir = load_params()
Get tensors from loaded_graph using the function get_tensor_by_name(). Get the tensors using the following names:
Return the tensors in the following tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)
hide_code
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_tensor = loaded_graph.get_tensor_by_name('input:0')
initial_state_tensor = loaded_graph.get_tensor_by_name('initial_state:0')
final_state_tensor = loaded_graph.get_tensor_by_name('final_state:0')
probs_tensor = loaded_graph.get_tensor_by_name('probs:0')
return input_tensor, initial_state_tensor, final_state_tensor, probs_tensor
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_get_tensors(get_tensors)
Implement the pick_word() function to select the next word using probabilities.
hide_code
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 int_to_vocab[np.argmax(probabilities)]
"""DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE"""
test_pick_word(pick_word)
This will generate the TV script for you. Set gen_length to the length of TV script you want to generate.
hide_code
gen_length = 200
# homer_simpson, moe_szyslak, or bart_simpson
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)
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.
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.