def train(self):
params = self.params
batches = mini_batches(X_added_code=self.added_code,
X_removed_code=self.removed_code,
Y=self.msgs,
mini_batch_size=params.batch_size)
msg_labels_shape = self.msgs.shape
self._clean_unused_variables()
params.save_dir = os.path.join(
params.save_dir,
datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S'))
params.vocab_code = len(self.dict_code)
if len(msg_labels_shape) == 1:
params.class_num = 1
else:
params.class_num = msg_labels_shape[1]
# Device configuration
params.device = torch.device('cuda' if params.cuda else 'cpu')
model = HierachicalRNN(args=params)
if params.cuda:
model = model.cuda()
optimizer = torch.optim.Adam(model.parameters(),
lr=params.l2_reg_lambda)
criterion = nn.BCEWithLogitsLoss()
for epoch in range(1, params.num_epochs + 1):
total_loss = 0
for i, (batch) in enumerate(tqdm(batches)):
# reset the hidden state of hierarchical attention model
state_word = model.init_hidden_word()
state_sent = model.init_hidden_sent()
state_hunk = model.init_hidden_hunk()
pad_added_code, pad_removed_code, labels = batch
labels = torch.cuda.FloatTensor(labels)
optimizer.zero_grad()
predict = model.forward(pad_added_code, pad_removed_code,
state_hunk, state_sent, state_word)
loss = criterion(predict, labels)
loss.backward()
total_loss += loss.detach()
optimizer.step()
print('Training: Epoch %i / %i -- Total loss: %f' %
(epoch, params.num_epochs, total_loss))
save(model, params.save_dir, 'epoch', epoch)
def train_model(data, params):
pad_added_code, pad_removed_code, pad_msg_labels, dict_msg, dict_code = data
batches = mini_batches(X_added_code=pad_added_code,
X_removed_code=pad_removed_code,
Y=pad_msg_labels,
mini_batch_size=params.batch_size)
torch.cuda.set_device(-1)
params.cuda = (not params.no_cuda) and torch.cuda.is_available()
del params.no_cuda
params.save_dir = os.path.join(
params.save_dir,
datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S'))
params.vocab_code = len(dict_code)
if len(pad_msg_labels.shape) == 1:
params.class_num = 1
else:
params.class_num = pad_msg_labels.shape[1]
# Device configuration
params.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
model = HierachicalRNN(args=params)
if torch.cuda.is_available():
model = model.cuda()
optimizer = torch.optim.Adam(model.parameters(), lr=params.l2_reg_lambda)
criterion = nn.BCEWithLogitsLoss()
batches = batches[:10]
for epoch in range(1, params.num_epochs + 1):
total_loss = 0
for i, (batch) in enumerate(tqdm(batches)):
# reset the hidden state of hierarchical attention model
state_word = model.init_hidden_word()
state_sent = model.init_hidden_sent()
state_hunk = model.init_hidden_hunk()
pad_added_code, pad_removed_code, labels = batch
labels = torch.cuda.FloatTensor(labels)
optimizer.zero_grad()
predict = model.forward(pad_added_code, pad_removed_code,
state_hunk, state_sent, state_word)
loss = criterion(predict, labels)
loss.backward()
total_loss += loss
optimizer.step()
print('Training: Epoch %i / %i -- Total loss: %f' %
(epoch, params.num_epochs, total_loss))
save(model, params.save_dir, 'epoch', epoch)
def predict(self):
params = self.params
batches = mini_batches(X_added_code=self.added_code,
X_removed_code=self.removed_code,
Y=self.msgs,
mini_batch_size=params.batch_size,
shuffled=False)
msg_labels_shape = self.msgs.shape
params.vocab_code = len(self.dict_code)
if len(msg_labels_shape) == 1:
params.class_num = 1
else:
params.class_num = msg_labels_shape[1]
# Device configuration
params.device = torch.device('cuda' if params.cuda else 'cpu')
model = HierachicalRNN(args=params)
model.load_state_dict(torch.load(params.load_model))
if params.cuda:
model = model.cuda()
model.eval(
) # eval mode (batchnorm uses moving mean/variance instead of mini-batch mean/variance)
commit_ftrs = list()
with torch.no_grad():
for i, (batch) in enumerate(tqdm(batches)):
state_word = model.init_hidden_word()
state_sent = model.init_hidden_sent()
state_hunk = model.init_hidden_hunk()
pad_added_code, pad_removed_code, labels = batch
labels = torch.cuda.FloatTensor(labels)
commit_ftr = model.forward_commit_embeds_diff(
pad_added_code, pad_removed_code, state_hunk, state_sent,
state_word)
commit_ftrs.append(commit_ftr)
commit_ftrs = torch.cat(commit_ftrs).cpu().detach().numpy()
pickle.dump(commit_ftrs, open(params.name, 'wb'))