def main(args):
# parse args
args = parse_args(args)
# prepare data
if args['prep_data']:
print('\n ====== preparing data ====== \n')
for i in range(1, 7):
print(' TASK #{}\n'.format(i))
prepare_data(args, task_id=i)
sys.exit()
##################################################################
# 데이터 준비가 아니면 read data and metadata from pickled files
##################################################################
with open(P_DATA_DIR + str(args['task_id']) + '.metadata.pkl', 'rb') as f:
metadata = pkl.load(f)
with open(P_DATA_DIR + str(args['task_id']) + '.data.pkl', 'rb') as f:
data_ = pkl.load(f)
# read content of data and metadata
candidates = data_['candidates']
candid2idx, idx2candid = metadata['candid2idx'], metadata['idx2candid']
# read train, test, val data
train, test, val = data_['train'], data_['test'], data_['val']
# get more required information from metadata
sentence_size = metadata['sentence_size']
w2idx = metadata['w2idx']
idx2w = metadata['idx2w']
memory_size = metadata['memory_size']
vocab_size = metadata['vocab_size']
n_cand = metadata['n_cand']
candidate_sentence_size = metadata['candidate_sentence_size']
# 후보 response들의 백터화
candidates_vec = data_utils.vectorize_candidates(candidates, w2idx,
candidate_sentence_size)
# create model - memn2n
model = memn2n.MemN2NDialog(batch_size=BATCH_SIZE,
vocab_size=vocab_size,
candidates_size=n_cand,
sentence_size=sentence_size,
embedding_size=20,
candidates_vec=candidates_vec,
hops=3)
train, val, test, batches = data_utils.get_batches(train,
val,
test,
metadata,
batch_size=BATCH_SIZE)
# training은 여기서 실행 된다.
if args['train']:
epochs = args['epochs']
eval_interval = args['eval_interval']
print('\n>>> Training started...\n')
# write log to file
log_handle = open('log/' + args['log_file'], 'w')
cost_total = 0.
for i in range(epochs + 1):
for start, end in batches:
s = train['s'][start:end]
q = train['q'][start:end]
a = train['a'][start:end]
cost_total += model.batch_fit(s, q, a)
if i % eval_interval == 0 and i:
train_preds = batch_predict(model,
train['s'],
train['q'],
len(train['s']),
batch_size=BATCH_SIZE)
val_preds = batch_predict(model,
val['s'],
val['q'],
len(val['s']),
batch_size=BATCH_SIZE)
train_acc = metrics.accuracy_score(np.array(train_preds),
train['a'])
val_acc = metrics.accuracy_score(val_preds, val['a'])
print(
'Epoch[{}] : <Accuracy>\n\ttraining : {} \n\tvalidation : {}'
.format(i, train_acc, val_acc))
log_handle.write('{} {} {} {}\n'.format(
i, train_acc, val_acc,
cost_total / (eval_interval * len(batches))))
cost_total = 0.
model.saver.save(
model._sess,
CKPT_DIR + '{}/memn2n_model.ckpt'.format(args['task_id']),
global_step=i)
log_handle.close()
# inference
else:
# restore checkpoint
ckpt = tf.train.get_checkpoint_state(CKPT_DIR + str(args['task_id']))
if ckpt and ckpt.model_checkpoint_path:
print('\n>> restoring checkpoint from', ckpt.model_checkpoint_path)
model.saver.restore(model._sess, ckpt.model_checkpoint_path)
isess = InteractiveSession(model, idx2candid, w2idx, n_cand,
memory_size)
if args['infer']:
query = ''
while query != 'exit':
query = input('>> ')
print('>> ' + isess.reply(query))
def main(args):
# parse args
args = parse_args(args)
# prepare data
if args['prep_data']:
print('\n>> Preparing Data\n')
for i in range(1,7):
print(' TASK#{}\n'.format(i))
prepare_data(args, task_id=i)
sys.exit()
# ELSE
# read data and metadata from pickled files
with open(P_DATA_DIR + str(args['task_id']) + '.metadata.pkl', 'rb') as f:
metadata = pkl.load(f)
with open(P_DATA_DIR + str(args['task_id']) + '.data.pkl', 'rb') as f:
data_ = pkl.load(f)
# read content of data and metadata
candidates = data_['candidates']
candid2idx, idx2candid = metadata['candid2idx'], metadata['idx2candid']
# get train/test/val data
train, test, val = data_['train'], data_['test'], data_['val']
# gather more information from metadata
sentence_size = metadata['sentence_size']
w2idx = metadata['w2idx']
idx2w = metadata['idx2w']
memory_size = metadata['memory_size']
vocab_size = metadata['vocab_size']
n_cand = metadata['n_cand']
candidate_sentence_size = metadata['candidate_sentence_size']
# vectorize candidates
candidates_vec = data_utils.vectorize_candidates(candidates, w2idx, candidate_sentence_size)
###
# create model
#model = model['memn2n']( # why?
model = memn2n.MemN2NDialog(
batch_size= BATCH_SIZE,
vocab_size= vocab_size,
candidates_size= n_cand,
sentence_size= sentence_size,
embedding_size= 20,
candidates_vec= candidates_vec,
hops= 3
)
# gather data in batches
train, val, test, batches = data_utils.get_batches(train, val, test, metadata, batch_size=BATCH_SIZE)
if args['train']:
# training starts here
epochs = args['epochs']
eval_interval = args['eval_interval']
#
# training and evaluation loop
print('\n>> Training started!\n')
# write log to file
log_handle = open('log/' + args['log_file'], 'w')
cost_total = 0.
#best_validation_accuracy = 0.
for i in range(epochs+1):
for start, end in batches:
s = train['s'][start:end]
q = train['q'][start:end]
a = train['a'][start:end]
cost_total += model.batch_fit(s, q, a)
if i%eval_interval == 0 and i:
train_preds = batch_predict(model, train['s'], train['q'], len(train['s']), batch_size=BATCH_SIZE)
val_preds = batch_predict(model, val['s'], val['q'], len(val['s']), batch_size=BATCH_SIZE)
train_acc = metrics.accuracy_score(np.array(train_preds), train['a'])
val_acc = metrics.accuracy_score(val_preds, val['a'])
print('Epoch[{}] : <ACCURACY>\n\ttraining : {} \n\tvalidation : {}'.
format(i, train_acc, val_acc))
log_handle.write('{} {} {} {}\n'.format(i, train_acc, val_acc,
cost_total/(eval_interval*len(batches))))
cost_total = 0. # empty cost
#
# save the best model, to disk
#if val_acc > best_validation_accuracy:
#best_validation_accuracy = val_acc
model.saver.save(model._sess, CKPT_DIR + '{}/memn2n_model.ckpt'.format(args['task_id']),
global_step=i)
model.saver.save(model._sess, CKPT_DIR + '{}/memn2n_model.ckpt_final'.format(args['task_id']),
global_step=i)
# close file
log_handle.close()
else: # inference
###
print("launching interative dialog session")
#print("model's session variables" + model._sess)
ckpt = tf.train.get_checkpoint_state(CKPT_DIR + '{}'.format(args['task_id']))
if ckpt and ckpt.model_checkpoint_path:
print('\n>> restoring checkpoint from', ckpt.model_checkpoint_path)
model.saver.restore(model._sess, ckpt.model_checkpoint_path)
ds = interactive_session(task_id=args['task_id'])
ds.interactive_dialog(task_id=args['task_id'])
'''