train.py

import tensorflow as tf

import numpy as np
import os
import time
import json

#
#             CONFIGURATION VARIABLES
#

# The embedding dimension
#embedding_dim = 256

# Number of RNN units
#rnn_units = 1024

#path_to_file = "data.txt"

# Batch size
#BATCH_SIZE = 64

# Buffer size to shuffle the dataset
# (TF data is designed to work with possibly infinite sequences,
# so it doesn't attempt to shuffle the entire sequence in memory. Instead,
# it maintains a buffer in which it shuffles elements).
#BUFFER_SIZE = 10000

# The maximum length sentence we want for a single input in characters
#seq_length = 100


with open("settings.json", 'r') as i:
  j = json.loads(i.read())

  embedding_dim = j['embedding_dim']
  rnn_units = j['rnn_units']
  path_to_file = j['path_to_file']
  BATCH_SIZE = j['BATCH_SIZE']
  BUFFER_SIZE = j['BUFFER_SIZE']
  seq_length = j['seq_length']
  EPOCHS = j['EPOCHS']


#
#    DON'T TOUCH
#

# Read, then decode for py2 compat.
text = open(path_to_file, 'rb').read().decode(encoding='utf-8')

vocab = sorted(set(text))

# Creating a mapping from unique characters to indices
char2idx = {u:i for i, u in enumerate(vocab)}
idx2char = np.array(vocab)

text_as_int = np.array([char2idx[c] for c in text])


examples_per_epoch = len(text)//(seq_length+1)

# Create training examples / targets
char_dataset = tf.data.Dataset.from_tensor_slices(text_as_int)

sequences = char_dataset.batch(seq_length+1, drop_remainder=True)

def split_input_target(chunk):
  input_text = chunk[:-1]
  target_text = chunk[1:]
  return input_text, target_text

dataset = sequences.map(split_input_target)

dataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)

# Length of the vocabulary in chars
vocab_size = len(vocab)



def build_model(vocab_size, embedding_dim, rnn_units, batch_size):
  model = tf.keras.Sequential([
    tf.keras.layers.Embedding(vocab_size, embedding_dim,
                              batch_input_shape=[batch_size, None]),
    tf.keras.layers.GRU(rnn_units,
                        return_sequences=True,
                        stateful=True,
                        recurrent_initializer='glorot_uniform'),
    tf.keras.layers.Dense(vocab_size)
  ])
  return model


checkpoint_dir = './training_checkpoints'
# Loads weight
try:
  tf.train.latest_checkpoint(checkpoint_dir)

  model = build_model(vocab_size, embedding_dim, rnn_units, batch_size=BATCH_SIZE)

  model.load_weights(tf.train.latest_checkpoint(checkpoint_dir))

  model.build(tf.TensorShape([1, None]))

except:
  model = build_model(
    vocab_size = len(vocab),
    embedding_dim=embedding_dim,
    rnn_units=rnn_units,
    batch_size=BATCH_SIZE)

def loss(labels, logits):
  return tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)


model.compile(optimizer='adam', loss=loss)
 
if not os.path.isdir("training_checkpoints"):
    os.mkdir("training_checkpoints")

# Directory where the checkpoints will be saved
checkpoint_dir = './training_checkpoints'
# Name of the checkpoint files
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt_{epoch}")

checkpoint_callback=tf.keras.callbacks.ModelCheckpoint(
    filepath=checkpoint_prefix,
    save_weights_only=True)

history = model.fit(dataset, epochs=EPOCHS, callbacks=[checkpoint_callback])