def train_RNNLM(_run):
# maintain consistency between sacred config and experiment config
config = ExperimentConfig(**_run.config)
experiment_dump_path = os.path.join(experiment_path, str(_run._id), "tf_dump")
if not os.path.exists(experiment_dump_path):
os.makedirs(experiment_dump_path)
if config.debug:
print("Running in debug mode...")
with tf.Graph().as_default():
# set random seed before the graph is built
tf.set_random_seed(config.tf_random_seed)
model = RNNLM_Model(config, load_corpus=True)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
with tf.Session(config=tf_config) as session:
best_val_pp = float('inf')
best_val_epoch = 0
tf.summary.FileWriter(os.path.join(train_path, "TensorLog"), session.graph)
session.run(init)
for epoch in range(config.max_epochs):
print('Epoch {}'.format(epoch))
start = time.time()
train_pp = model.run_epoch(
session, model.encoded_train,
train_op=model.train_step)
print('Training perplexity: {}'.format(train_pp))
print('Total Training time: {}'.format(time.time() - start))
valid_pp = model.run_epoch(session, model.encoded_dev)
print('Validation perplexity: {}'.format(valid_pp))
if valid_pp < best_val_pp:
best_val_pp = valid_pp
best_val_epoch = epoch
saver.save(session, os.path.join(experiment_dump_path, 'rnnlm.weights'))
if epoch - best_val_epoch > config.early_stopping:
break
print('Total time: {}'.format(time.time() - start))
test_pp = model.run_epoch(session, model.encoded_test)
print('Test perplexity: {}'.format(test_pp))
def dump_trained_weights(experiment, verbose):
config = get_configs(experiment)
# Still need to load the model to build graph
# Graph is not saved
RNNLM_Model(config)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as session:
session.run(init)
saver.restore(
session,
os.path.join(experiment_path, str(experiment), "tf_dump",
'rnnlm.weights'))
dump_vars = [
'HMi', 'HMf', 'HMo', 'HMo', 'HMg', 'IMi', 'IMf', 'IMo', 'IMg',
'LM', 'bi', 'bf', 'bo', 'bg', 'b2'
]
if config.share_embedding:
dump_vars += ['PM']
else:
dump_vars += ['UM']
if config.V_table:
dump_vars.remove('LM')
for i, seg in enumerate(config.embedding_seg):
if i != 0:
dump_vars += ['VT{}'.format(i)]
dump_vars += ['LM{}'.format(i)]
weight_dict = tf_weights_to_np_weights_dict(session, dump_vars)
if config.D_softmax:
# instead save the full patched embedding, split each block in "LM" into list of matrices
blocks = []
col_s = 0
for size, s, e in config.embedding_seg:
if e is None:
e = weight_dict['LM'].shape[0]
blocks.append(weight_dict['LM'][s:e, col_s:col_s + size])
col_s += size
weight_dict['LM'] = blocks
weight_dump_dir = os.path.join(experiment_path, str(experiment),
"weights")
dump_weights(weight_dict, weight_dump_dir, verbose)
def auto_generate_sentence(experiment=1):
gen_config = get_configs(experiment)
gen_config.batch_size = gen_config.num_steps = 1
gen_model = RNNLM_Model(gen_config)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as session:
session.run(init)
saver.restore(
session,
os.path.join(experiment_path, str(experiment), "tf_dump",
'rnnlm.weights'))
starting_text = '<eos>'
while starting_text:
sen = generate_sentence(session,
gen_model,
gen_config,
starting_text=starting_text,
temp=1.0)
print(' '.join([w.split('/')[0] for w in sen]))
starting_text = input('> ')