In this project, you’re going to take a peek into the realm of neural network machine translation. You’ll be training a sequence to sequence model on a dataset of English and French sentences that can translate new sentences from English to French.
Since translating the whole language of English to French will take lots of time to train, we have provided you with a small portion of the English corpus.
In [8]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
import problem_unittests as tests
source_path = 'data/small_vocab_en'
target_path = 'data/small_vocab_fr'
source_text = helper.load_data(source_path)
target_text = helper.load_data(target_path)
In [2]:
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 source_text.split()})))
sentences = source_text.split('\n')
word_counts = [len(sentence.split()) for sentence in sentences]
print('Number of sentences: {}'.format(len(sentences)))
print('Average number of words in a sentence: {}'.format(np.average(word_counts)))
print()
print('English sentences {} to {}:'.format(*view_sentence_range))
print('\n'.join(source_text.split('\n')[view_sentence_range[0]:view_sentence_range[1]]))
print()
print('French sentences {} to {}:'.format(*view_sentence_range))
print('\n'.join(target_text.split('\n')[view_sentence_range[0]:view_sentence_range[1]]))
As you did with other RNNs, you must turn the text into a number so the computer can understand it. In the function text_to_ids(), you'll turn source_text and target_text from words to ids. However, you need to add the <EOS> word id at the end of each sentence from target_text. This will help the neural network predict when the sentence should end.
You can get the <EOS> word id by doing:
target_vocab_to_int['<EOS>']
You can get other word ids using source_vocab_to_int and target_vocab_to_int.
In [3]:
def single_text_to_ids(text, vocab_to_int, add_EOS):
id_text = []
for sentence in text.split('\n'):
id_sentence = []
for word in sentence.split():
id_sentence.append(vocab_to_int[word])
if add_EOS:
id_sentence.append(vocab_to_int['<EOS>'])
#print(sentence)
#print(id_sentence)
id_text.append(id_sentence)
#print(id_text)
return id_text
def text_to_ids(source_text, target_text, source_vocab_to_int, target_vocab_to_int):
"""
Convert source and target text to proper word ids
:param source_text: String that contains all the source text.
:param target_text: String that contains all the target text.
:param source_vocab_to_int: Dictionary to go from the source words to an id
:param target_vocab_to_int: Dictionary to go from the target words to an id
:return: A tuple of lists (source_id_text, target_id_text)
"""
# TODO: Implement Function
#print(source_text)
#print(target_text)
#print(source_vocab_to_int)
#print(target_vocab_to_int)
source_id_text = single_text_to_ids(source_text, source_vocab_to_int, False)
target_id_text = single_text_to_ids(target_text, target_vocab_to_int, True)
return source_id_text, target_id_text
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_text_to_ids(text_to_ids)
In [4]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
helper.preprocess_and_save_data(source_path, target_path, text_to_ids)
In [9]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np
import helper
(source_int_text, target_int_text), (source_vocab_to_int, target_vocab_to_int), _ = helper.load_preprocess()
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__) in [LooseVersion('1.0.0'), LooseVersion('1.0.1')], 'This project requires TensorFlow version 1.0 You are using {}'.format(tf.__version__)
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()))
You'll build the components necessary to build a Sequence-to-Sequence model by implementing the following functions below:
model_inputsprocess_decoding_inputencoding_layerdecoding_layer_traindecoding_layer_inferdecoding_layerseq2seq_modelImplement the model_inputs() function to create TF Placeholders for the Neural Network. It should create the following placeholders:
Return the placeholders in the following the tuple (Input, Targets, Learing Rate, Keep Probability)
In [11]:
def model_inputs():
"""
Create TF Placeholders for input, targets, and learning rate.
:return: Tuple (input, targets, learning rate, keep probability)
"""
# TODO: Implement Function
input = 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')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
return input, targets, learning_rate, keep_prob
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)
In [12]:
def process_decoding_input(target_data, target_vocab_to_int, batch_size):
"""
Preprocess target data for dencoding
:param target_data: Target Placehoder
:param target_vocab_to_int: Dictionary to go from the target words to an id
:param batch_size: Batch Size
:return: Preprocessed target data
"""
# TODO: Implement Function
ending = tf.strided_slice(target_data, [0, 0], [batch_size, -1], [1, 1])
dec_input = tf.concat([tf.fill([batch_size, 1], target_vocab_to_int['<GO>']), ending], 1)
return dec_input
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_process_decoding_input(process_decoding_input)
Implement encoding_layer() to create a Encoder RNN layer using tf.nn.dynamic_rnn().
In [13]:
def encoding_layer(rnn_inputs, rnn_size, num_layers, keep_prob):
"""
Create encoding layer
:param rnn_inputs: Inputs for the RNN
:param rnn_size: RNN Size
:param num_layers: Number of layers
:param keep_prob: Dropout keep probability
:return: RNN state
"""
# TODO: Implement Function
lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
lstm = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob)
enc_cell = tf.contrib.rnn.MultiRNNCell([lstm] * num_layers)
_, enc_state = tf.nn.dynamic_rnn(enc_cell, rnn_inputs, dtype=tf.float32)
return enc_state
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_encoding_layer(encoding_layer)
Create training logits using tf.contrib.seq2seq.simple_decoder_fn_train() and tf.contrib.seq2seq.dynamic_rnn_decoder(). Apply the output_fn to the tf.contrib.seq2seq.dynamic_rnn_decoder() outputs.
In [14]:
def decoding_layer_train(encoder_state, dec_cell, dec_embed_input, sequence_length, decoding_scope,
output_fn, keep_prob):
"""
Create a decoding layer for training
:param encoder_state: Encoder State
:param dec_cell: Decoder RNN Cell
:param dec_embed_input: Decoder embedded input
:param sequence_length: Sequence Length
:param decoding_scope: TenorFlow Variable Scope for decoding
:param output_fn: Function to apply the output layer
:param keep_prob: Dropout keep probability
:return: Train Logits
"""
# TODO: Implement Function
#with tf.variable_scope("decoding") as decoding_scope:
# Training Decoder
train_decoder_fn = tf.contrib.seq2seq.simple_decoder_fn_train(encoder_state)
dec_cell = tf.contrib.rnn.DropoutWrapper(dec_cell, output_keep_prob=keep_prob)
train_pred, _, _ = tf.contrib.seq2seq.dynamic_rnn_decoder(
dec_cell, train_decoder_fn, dec_embed_input, sequence_length, scope=decoding_scope)
# Apply output function
train_logits = output_fn(train_pred)
return train_logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_decoding_layer_train(decoding_layer_train)
Create inference logits using tf.contrib.seq2seq.simple_decoder_fn_inference() and tf.contrib.seq2seq.dynamic_rnn_decoder().
In [15]:
def decoding_layer_infer(encoder_state, dec_cell, dec_embeddings, start_of_sequence_id, end_of_sequence_id,
maximum_length, vocab_size, decoding_scope, output_fn, keep_prob):
"""
Create a decoding layer for inference
:param encoder_state: Encoder state
:param dec_cell: Decoder RNN Cell
:param dec_embeddings: Decoder embeddings
:param start_of_sequence_id: GO ID
:param end_of_sequence_id: EOS Id
:param maximum_length: The maximum allowed time steps to decode
:param vocab_size: Size of vocabulary
:param decoding_scope: TensorFlow Variable Scope for decoding
:param output_fn: Function to apply the output layer
:param keep_prob: Dropout keep probability
:return: Inference Logits
"""
# TODO: Implement Function
# Inference Decoder
infer_decoder_fn = tf.contrib.seq2seq.simple_decoder_fn_inference(
output_fn, encoder_state, dec_embeddings, start_of_sequence_id, end_of_sequence_id,
maximum_length, vocab_size)
inference_logits, _, _ = tf.contrib.seq2seq.dynamic_rnn_decoder(dec_cell, infer_decoder_fn, scope=decoding_scope)
return inference_logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_decoding_layer_infer(decoding_layer_infer)
Implement decoding_layer() to create a Decoder RNN layer.
rnn_size and num_layers.lambda to transform it's input, logits, to class logits.decoding_layer_train(encoder_state, dec_cell, dec_embed_input, sequence_length, decoding_scope, output_fn, keep_prob) function to get the training logits.decoding_layer_infer(encoder_state, dec_cell, dec_embeddings, start_of_sequence_id, end_of_sequence_id, maximum_length, vocab_size, decoding_scope, output_fn, keep_prob) function to get the inference logits.Note: You'll need to use tf.variable_scope to share variables between training and inference.
In [27]:
def decoding_layer(dec_embed_input, dec_embeddings, encoder_state, vocab_size, sequence_length, rnn_size,
num_layers, target_vocab_to_int, keep_prob):
"""
Create decoding layer
:param dec_embed_input: Decoder embedded input
:param dec_embeddings: Decoder embeddings
:param encoder_state: The encoded state
:param vocab_size: Size of vocabulary
:param sequence_length: Sequence Length
:param rnn_size: RNN Size
:param num_layers: Number of layers
:param target_vocab_to_int: Dictionary to go from the target words to an id
:param keep_prob: Dropout keep probability
:return: Tuple of (Training Logits, Inference Logits)
"""
# TODO: Implement Function
# Decoder RNNs
lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
lstm = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob)
dec_cell = tf.contrib.rnn.MultiRNNCell([lstm] * num_layers)
with tf.variable_scope("decoding") as decoding_scope:
# Output Layer
output_fn = lambda x: tf.contrib.layers.fully_connected(x, vocab_size, None, scope=decoding_scope)
#with tf.variable_scope("decoding") as decoding_scope:
train_logits = decoding_layer_train(encoder_state, dec_cell, dec_embed_input, sequence_length, decoding_scope, output_fn, keep_prob)
with tf.variable_scope("decoding", reuse=True) as decoding_scope:
start_of_sequence_id = target_vocab_to_int['<GO>']
end_of_sequence_id = target_vocab_to_int['<EOS>']
maximum_length = sequence_length - 1
inference_logits = decoding_layer_infer(encoder_state, dec_cell, dec_embeddings, start_of_sequence_id, end_of_sequence_id, maximum_length, vocab_size, decoding_scope, output_fn, keep_prob)
return train_logits, inference_logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_decoding_layer(decoding_layer)
Apply the functions you implemented above to:
encoding_layer(rnn_inputs, rnn_size, num_layers, keep_prob).process_decoding_input(target_data, target_vocab_to_int, batch_size) function.decoding_layer(dec_embed_input, dec_embeddings, encoder_state, vocab_size, sequence_length, rnn_size, num_layers, target_vocab_to_int, keep_prob).
In [17]:
def seq2seq_model(input_data, target_data, keep_prob, batch_size, sequence_length, source_vocab_size, target_vocab_size,
enc_embedding_size, dec_embedding_size, rnn_size, num_layers, target_vocab_to_int):
"""
Build the Sequence-to-Sequence part of the neural network
:param input_data: Input placeholder
:param target_data: Target placeholder
:param keep_prob: Dropout keep probability placeholder
:param batch_size: Batch Size
:param sequence_length: Sequence Length
:param source_vocab_size: Source vocabulary size
:param target_vocab_size: Target vocabulary size
:param enc_embedding_size: Decoder embedding size
:param dec_embedding_size: Encoder embedding size
:param rnn_size: RNN Size
:param num_layers: Number of layers
:param target_vocab_to_int: Dictionary to go from the target words to an id
:return: Tuple of (Training Logits, Inference Logits)
"""
# TODO: Implement Function
#Apply embedding to the input data for the encoder.
enc_embed_input = tf.contrib.layers.embed_sequence(input_data, source_vocab_size, enc_embedding_size)
#Encode the input using your encoding_layer(rnn_inputs, rnn_size, num_layers, keep_prob).
enc_state = encoding_layer(enc_embed_input, rnn_size, num_layers, keep_prob)
#Process target data using your process_decoding_input(target_data, target_vocab_to_int, batch_size) function.
dec_input = process_decoding_input(target_data, target_vocab_to_int, batch_size)
#Apply embedding to the target data for the decoder.
dec_embeddings = tf.Variable(tf.random_uniform([target_vocab_size, dec_embedding_size]))
dec_embed_input = tf.nn.embedding_lookup(dec_embeddings, dec_input)
#Decode the encoded input using your decoding_layer(dec_embed_input, dec_embeddings, encoder_state, vocab_size, sequence_length, rnn_size, num_layers, target_vocab_to_int, keep_prob).
train_logits, inference_logits = decoding_layer(dec_embed_input, dec_embeddings, enc_state, target_vocab_size, sequence_length, rnn_size, num_layers, target_vocab_to_int, keep_prob)
return train_logits, inference_logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_seq2seq_model(seq2seq_model)
Tune the following parameters:
epochs to the number of epochs.batch_size to the batch size.rnn_size to the size of the RNNs.num_layers to the number of layers.encoding_embedding_size to the size of the embedding for the encoder.decoding_embedding_size to the size of the embedding for the decoder.learning_rate to the learning rate.keep_probability to the Dropout keep probability
In [18]:
# Number of Epochs
epochs = 20
# Batch Size
batch_size = 512
# RNN Size
rnn_size = 128
# Number of Layers
num_layers = 2
# Embedding Size
encoding_embedding_size = 200
decoding_embedding_size = 200
# Learning Rate
learning_rate = 0.001
# Dropout Keep Probability
keep_probability = 0.5
In [19]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
save_path = 'checkpoints/dev'
(source_int_text, target_int_text), (source_vocab_to_int, target_vocab_to_int), _ = helper.load_preprocess()
max_source_sentence_length = max([len(sentence) for sentence in source_int_text])
train_graph = tf.Graph()
with train_graph.as_default():
input_data, targets, lr, keep_prob = model_inputs()
sequence_length = tf.placeholder_with_default(max_source_sentence_length, None, name='sequence_length')
input_shape = tf.shape(input_data)
train_logits, inference_logits = seq2seq_model(
tf.reverse(input_data, [-1]), targets, keep_prob, batch_size, sequence_length, len(source_vocab_to_int), len(target_vocab_to_int),
encoding_embedding_size, decoding_embedding_size, rnn_size, num_layers, target_vocab_to_int)
tf.identity(inference_logits, 'logits')
with tf.name_scope("optimization"):
# Loss function
cost = tf.contrib.seq2seq.sequence_loss(
train_logits,
targets,
tf.ones([input_shape[0], sequence_length]))
# 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)
In [20]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import time
def get_accuracy(target, logits):
"""
Calculate accuracy
"""
max_seq = max(target.shape[1], logits.shape[1])
if max_seq - target.shape[1]:
target = np.pad(
target,
[(0,0),(0,max_seq - target.shape[1])],
'constant')
if max_seq - logits.shape[1]:
logits = np.pad(
logits,
[(0,0),(0,max_seq - logits.shape[1]), (0,0)],
'constant')
return np.mean(np.equal(target, np.argmax(logits, 2)))
train_source = source_int_text[batch_size:]
train_target = target_int_text[batch_size:]
valid_source = helper.pad_sentence_batch(source_int_text[:batch_size])
valid_target = helper.pad_sentence_batch(target_int_text[:batch_size])
with tf.Session(graph=train_graph) as sess:
sess.run(tf.global_variables_initializer())
for epoch_i in range(epochs):
for batch_i, (source_batch, target_batch) in enumerate(
helper.batch_data(train_source, train_target, batch_size)):
start_time = time.time()
_, loss = sess.run(
[train_op, cost],
{input_data: source_batch,
targets: target_batch,
lr: learning_rate,
sequence_length: target_batch.shape[1],
keep_prob: keep_probability})
batch_train_logits = sess.run(
inference_logits,
{input_data: source_batch, keep_prob: 1.0})
batch_valid_logits = sess.run(
inference_logits,
{input_data: valid_source, keep_prob: 1.0})
train_acc = get_accuracy(target_batch, batch_train_logits)
valid_acc = get_accuracy(np.array(valid_target), batch_valid_logits)
end_time = time.time()
if batch_i % 10 == 0:
print('Epoch {:>3} Batch {:>4}/{} - Train Accuracy: {:>6.3f}, Validation Accuracy: {:>6.3f}, Loss: {:>6.3f}'
.format(epoch_i, batch_i, len(source_int_text) // batch_size, train_acc, valid_acc, loss))
# Save Model
saver = tf.train.Saver()
saver.save(sess, save_path)
print('Model Trained and Saved')
In [21]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Save parameters for checkpoint
helper.save_params(save_path)
In [22]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import tensorflow as tf
import numpy as np
import helper
import problem_unittests as tests
_, (source_vocab_to_int, target_vocab_to_int), (source_int_to_vocab, target_int_to_vocab) = helper.load_preprocess()
load_path = helper.load_params()
To feed a sentence into the model for translation, you first need to preprocess it. Implement the function sentence_to_seq() to preprocess new sentences.
vocab_to_int<UNK> word id.
In [23]:
def sentence_to_seq(sentence, vocab_to_int):
"""
Convert a sentence to a sequence of ids
:param sentence: String
:param vocab_to_int: Dictionary to go from the words to an id
:return: List of word ids
"""
# TODO: Implement Function
lower_sentence = sentence.lower()
id_seq = []
for word in lower_sentence.split():
id_seq.append(vocab_to_int.get(word, vocab_to_int['<UNK>']))
return id_seq
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_sentence_to_seq(sentence_to_seq)
In [24]:
translate_sentence = 'he saw a old yellow truck .'
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
translate_sentence = sentence_to_seq(translate_sentence, source_vocab_to_int)
loaded_graph = tf.Graph()
with tf.Session(graph=loaded_graph) as sess:
# Load saved model
loader = tf.train.import_meta_graph(load_path + '.meta')
loader.restore(sess, load_path)
input_data = loaded_graph.get_tensor_by_name('input:0')
logits = loaded_graph.get_tensor_by_name('logits:0')
keep_prob = loaded_graph.get_tensor_by_name('keep_prob:0')
translate_logits = sess.run(logits, {input_data: [translate_sentence], keep_prob: 1.0})[0]
print('Input')
print(' Word Ids: {}'.format([i for i in translate_sentence]))
print(' English Words: {}'.format([source_int_to_vocab[i] for i in translate_sentence]))
print('\nPrediction')
print(' Word Ids: {}'.format([i for i in np.argmax(translate_logits, 1)]))
print(' French Words: {}'.format([target_int_to_vocab[i] for i in np.argmax(translate_logits, 1)]))
You might notice that some sentences translate better than others. Since the dataset you're using only has a vocabulary of 227 English words of the thousands that you use, you're only going to see good results using these words. Additionally, the translations in this data set were made by Google translate, so the translations themselves aren't particularly good. (We apologize to the French speakers out there!) Thankfully, for this project, you don't need a perfect translation. However, if you want to create a better translation model, you'll need better data.
You can train on the WMT10 French-English corpus. This dataset has more vocabulary and richer in topics discussed. However, this will take you days to train, so make sure you've a GPU and the neural network is performing well on dataset we provided. Just make sure you play with the WMT10 corpus after you've submitted this project.
When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_language_translation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.