if config['cli_dialog']:
env.cli_dialog(
restore_path=restore_path,
log_file=os.path.expanduser(config['cli_dialog_file']),
vocabulary=vocabulary,
character_positions_in_vocabulary=cpiv,
batch_generator_class=BatchGenerator,
reset_state_after_model_answer=args.
reset_state_after_model_answer, # if True after bot answer hidden state is reset
append_logs=
False, # if False and log_file already exists logs are put in to log_file + `#[index]`
answer_len_limit=500., # max number of characters in bot answer
randomize=not config[
'do_not_randomize_hidden_state'], # if True model hidden state is initialized with random numbers
preprocess_f=preprocess_f,
postprocess_f=postprocess_f,
temperature=config['temperature'],
)
# print('before telegram method')
if config['telegram_bot']:
env.telegram(
kwargs_for_building=kwargs_for_model_building,
restore_path=restore_path,
vocabulary=vocabulary,
character_positions_in_vocabulary=cpiv,
batch_generator_class=BatchGenerator,
log_path=config['telegram_log_path'],
build=False,
temperature=config['temperature'],
)
vocabulary = create_vocabulary(text)
vocabulary_size = len(vocabulary)
env = Environment(Lstm, LstmBatchGenerator, vocabulary=vocabulary)
cpiv = get_positions_in_vocabulary(vocabulary)
kwargs_for_building = dict(
batch_size=64,
num_layers=2,
num_nodes=[300, 300],
num_output_layers=2,
num_output_nodes=[124],
vocabulary_size=vocabulary_size,
# dim_compressed=10,
num_unrollings=30,
init_parameter=3.,
regularization_rate=.00001,
regime='inference',
going_to_limit_memory=True)
add_feed = [{'placeholder': 'dropout', 'value': 0.9}]
valid_add_feed = [{'placeholder': 'dropout', 'value': 1.}]
env.telegram(kwargs_for_building,
'debugging_lstm_and_gru/first/checkpoints/10ZZXXZZZZZZZZZZZZ0',
'telegram/debug',
vocabulary,
cpiv,
LstmBatchGenerator,
additions_to_feed_dict=valid_add_feed)
