def build_training_and_verification_sets():
"""
Training and Verification Set Builder
Builds the training and verification datasets. Depending on the
configuration, this may be from the source files or from pickled
files.
"""
if not Config.Train.restore:
input_str = read_input()
create_examples(input_str)
# Character to integer map required during text generation
Config.export_character_to_integer_map()
# Export the training and verification data in case
# the previous setup will be trained on aga
Config.export_train_and_verification_data()
Config.word_count = len(input_str.split(" "))
else:
Config.import_character_to_integer_map()
Config.import_train_and_verification_data()
Config.dataset_size = Config.Train.size() + Config.Validation.size()
_print_basic_text_statistics()
feed_dict={
x: input_x,
seq_len: phrase_seq_len
})
pred_char_id = Config.DecisionEngine.function(sess, softmax_out)
pred_char = Config.Generate.int2char()[pred_char_id]
generated_text.append(pred_char)
if cur_seq_len == Config.sequence_length:
# Delete off the front of the list if it has reached the specified sequence length
del input_x[0][0]
else:
# Shave last dummy element off since fixed batch size
del input_x[0][Config.sequence_length - 1]
cur_seq_len += 1
input_x[0].insert(cur_seq_len - 1, pred_char_id)
Config.Generate.prev_char = pred_char
logging.info("Output Text: " + Config.Generate.seed_text +
"".join(generated_text))
if __name__ == "__main__":
Config.parse_args()
Config.import_character_to_integer_map()
Config.Generate.build_int2char()
Config.Generate.build_seed_x()
generate_text()