print('Target output sequences')
print(data_fr_out[:2])
dataset = tf.data.Dataset.from_tensor_slices(
(data_en, data_fr_in, data_fr_out))
dataset = dataset.shuffle(len(raw_data_en)).batch(BATCH_SIZE,
drop_remainder=True)
en_vocab_size = len(en_tokenizer.word_index) + 1
encoder = Encoder(en_vocab_size, EMBEDDING_SIZE, RNN_SIZE)
fr_vocab_size = len(fr_tokenizer.word_index) + 1
decoder = Decoder(fr_vocab_size, EMBEDDING_SIZE, RNN_SIZE, ATTENTION_FUNC)
initial_state = encoder.init_states(1)
encoder_outputs = encoder(tf.constant([[1]]), initial_state)
decoder_outputs = decoder(tf.constant([[1]]), encoder_outputs[1:],
encoder_outputs[0])
def predict(test_source_text=None):
if test_source_text is None:
test_source_text = raw_data_en[np.random.choice(len(raw_data_en))]
test_source_seq = en_tokenizer.texts_to_sequences([test_source_text])
if len(test_source_seq[0]) == 0:
dists.append(test_source_text)
return 0, test_source_text, test_source_text
en_initial_states = encoder.init_states(1)
dataset = tf.data.Dataset.from_tensor_slices(
(data_en, data_fr_in, data_fr_out))
dataset = dataset.shuffle(len(raw_data_en)).batch(BATCH_SIZE,
drop_remainder=True)
en_vocab_size = len(en_tokenizer.word_index) + 1
encoder = Encoder(en_vocab_size, EMBEDDING_SIZE, RNN_SIZE)
fr_vocab_size = len(fr_tokenizer.word_index) + 1
decoder = Decoder(fr_vocab_size, EMBEDDING_SIZE, RNN_SIZE, ATTENTION_FUNC)
# These lines can be used for debugging purpose
# Or can be seen as a way to build the models
initial_state = encoder.init_states(1)
encoder_outputs = encoder(tf.constant([[1]]), initial_state)
decoder_outputs = decoder(tf.constant([[1]]), encoder_outputs[1:],
encoder_outputs[0])
optimizer = tf.keras.optimizers.Adam(clipnorm=5.0)
@tf.function
def train_step(source_seq, target_seq_in, target_seq_out, en_initial_states):
loss = 0
with tf.GradientTape() as tape:
en_outputs = encoder(source_seq, en_initial_states)
en_states = en_outputs[1:]
de_state_h, de_state_c = en_states