def train(self):
train_loss = np.array([])
valid_loss = np.array([])
print("start train")
for epoch in range(self.epoch_num):
# train
print('epoch{0}'.format(epoch))
start = time.time()
self.model.train()
tmp_train_loss, _, _, _, _ = self._run(
mode='train', data_loader=self.train_data_loader)
train_loss = np.append(train_loss,
tmp_train_loss.cpu().clone().numpy())
self.model.eval()
with torch.no_grad():
tmp_valid_loss, est_source, est_mask, noisy_amp_spec, clean_amp_spec = self._run(
mode='validation', data_loader=self.valid_data_loader)
valid_loss = np.append(valid_loss,
tmp_valid_loss.cpu().clone().numpy())
if (epoch + 1) % 10 == 0:
plot_time = time.time()
est_source = taF.complex_norm(est_source)
show_TF_domein_result(train_loss, valid_loss,
noisy_amp_spec[0, :, :],
est_mask[0, :, :], est_source[0, :, :],
clean_amp_spec[0, :, :])
print('plot_time:', time.time() - plot_time)
torch.save(self.model.state_dict(),
self.save_path + 'u_net{0}.ckpt'.format(epoch + 1))
end = time.time()
print('----excute time: {0}'.format(end - start))
def train(self):
train_loss = np.array([])
print("start train")
for epoch in range(self.epoch_num):
# train
print('epoch{0}'.format(epoch))
start = time.time()
self.model.train()
tmp_train_loss, est_source, est_mask, mix_amp_spec, true_amp_spec = self._run(
mode='train', data_loader=self.train_data_loader)
train_loss = np.append(train_loss,
tmp_train_loss.cpu().clone().numpy())
if (epoch + 1) % 10 == 0:
plot_time = time.time()
show_TF_domein_result(train_loss, mix_amp_spec[0, :, :],
est_mask[0,
0, :, :], est_source[0,
0, :, :],
true_amp_spec[0, :, :])
print('plot_time:', time.time() - plot_time)
torch.save(self.model.state_dict(),
self.save_path + 'u_net{0}.ckpt'.format(epoch + 1))
end = time.time()
print('----excute time: {0}'.format(end - start))
def train(self):
train_loss = np.array([])
valid_loss = np.array([])
print("start train")
for epoch in range(self.epoch_num):
# train
print('epoch{0}'.format(epoch))
start = time.time()
self.model.train()
tmp_train_loss, _, _, _ = self._run(self.model, self.criterion, self.train_data_loader, self.train_batch_size, mode='train')
train_loss = np.append(train_loss, tmp_train_loss.cpu().clone().numpy())
# validation
self.model.eval()
with torch.no_grad():
tmp_valid_loss, est_source, est_mask, mix_amp_spec = self._run(self.model, self.criterion, self.valid_data_loader, self.valid_batch_size, mode='validation')
valid_loss = np.append(valid_loss, tmp_valid_loss.cpu().clone().numpy())
if (epoch + 1) % 10 == 0:
torch.save(self.model.state_dict(), self.save_path + 'u_net{0}.ckpt'.format(epoch + 1))
end = time.time()
print('----excute time: {0}'.format(end - start))
show_TF_domein_result(valid_loss, mix_amp_spec[0,:,:], est_mask[0,0,:,:], est_source[0,0,:,:])