def _get_batch_input(self, batch_data):
batch_input = batchify(batch_data, self.args.use_char_emb)
if self.use_cuda:
batch_input = [Variable(x.cuda(async=True)) for x in batch_input]
else:
batch_input = [Variable(x) for x in batch_input]
return batch_input
def _iter_data(self, data):
num_iter = (len(data) + self.batch_size - 1) // self.batch_size
for i in range(num_iter):
start_idx = i * self.batch_size
batch_data = data[start_idx:(start_idx + self.batch_size)]
batch_input = batchify(batch_data)
# Transfer to GPU
if self.use_cuda:
batch_input = [Variable(x.cuda(non_blocking=True)) for x in batch_input]
else:
batch_input = [Variable(x) for x in batch_input]
yield batch_input
def _iter_data(self, data):
num_iter = (len(data) + self.batch_size - 1) // self.batch_size
for i in range(num_iter - 1): #hack remove -1
start_idx = i * self.batch_size
batch_data = data[start_idx:(start_idx + self.batch_size)]
batch_input = batchify(batch_data)
# Transfer to GPU
#if self.use_cuda:
# batch_input = [Variable(x.cuda(async=True)) for x in batch_input]
#else:
#print ("BATCH",batch_input)
#batch_input = [Variable(x) for x in batch_input]
yield batch_input
def _iter_data(self, data):
choice_data, nli_data = data
num_iter = (len(nli_data) + self.batch_size - 1) // self.batch_size
for i in range(num_iter):
start_idx = i * self.batch_size
batch_data = nli_data[start_idx:(start_idx + self.batch_size)]
batch_input = batchify(choice_data, batch_data)
# Transfer to GPU
if self.use_cuda:
batch_input = [Variable(x.cuda(async=True)) for x in batch_input]
else:
batch_input = [Variable(x) for x in batch_input]
yield batch_input
def _iter_data(self, data):
num_iter = (len(data) + self.batch_size - 1) // self.batch_size
for i in range(num_iter):
start_idx = i * self.batch_size
batch_data = data[start_idx:(start_idx + self.batch_size)]
# convert a batch into tensors
# batch_input = batchify(batch_data)
batch_input = batchify(batch_data, self.elmo)
# Transfer to GPU
if self.use_cuda:
batch_input = [Variable(x.cuda(async=True)) for x in batch_input]
else:
batch_input = [Variable(x) for x in batch_input]
yield batch_input
### create model
model = LSTMClassifier(embedding_dim=embedding_dim,
hidden_dim=hidden_dim,
label_size=n_label,
batch_size=batch_size,
use_gpu=False,
dropout_emb=dropout_emb)
### data processing
train_data = load_data(processed_train)
dev_data = load_data(processed_dev)
num_iter = (len(train_data) + batch_size - 1) // batch_size
for i in range(num_iter):
p, p_mask, q, q_mask, c, c_mask, y = batchify(
train_data[i * batch_size:(i + 1) * batch_size])
# print("passage", p, "\n", len(p), type(p))
# print("passage mask", p_mask, "\n", len(p_mask), type(p_mask))
# print("choice", c, "\n", len(c), type(c))
# print("choice mask", c_mask, "\n", len(c_mask), type(c_mask))
# print("question ", q, "\n", len(q), type(q))
# print("question mask", q_mask, "\n", len(q_mask), type(q_mask))
break
for i in range(epochs):
print('Epoch %d...' % i)
if i == 0:
dev_acc = model.evaluate(dev_data)
print('Dev accuracy: %f' % dev_acc)
start_time = time.time()
