alb_list = None
raa_list = None
vza_list = None
sza_list = None
toz_list = None
rad = None
albwf = None
o3wf = None
#train_loader
for i, (features, radiances) in enumerate(train_loader):
print(i, features.shape, radiances.shape)
#if real_epoch == 0:
optimizer.zero_grad()
outputs = model(features)
loss = msre(outputs, radiances)
loss.backward()
optimizer.step()
#train_losses.append(loss.item())
train_losses = np.append(train_losses, loss.item())
#train_losses.append(loss.data)
#if real_epoch == 0:
# each batch is 128, print this for one of 10 batches
if (i * 128) % (128 * 10) == 0:
print(f'{i * 128} / ',
len(train_loader) * 128,
time.time() - timestamp, datetime.datetime.now())
print(loss.item())
timestamp = time.time()
vza_list = None
sza_list = None
toz_list = None
rad = None
albwf = None
o3wf = None
#train_loader
for i, (features, radiances) in enumerate(train_loader):
print(i, features.shape, radiances.shape)
#if epochs_done == 0:
optimizer_l300.zero_grad()
optimizer_s300.zero_grad()
outputs_s300 = model_s300(features)
outputs_l300 = model_l300(features)
loss_s300 = msre(outputs_s300, radiances[:, :660])
loss_l300 = msre(outputs_l300, radiances[:, 660:])
loss_s300.backward()
loss_l300.backward()
optimizer_l300.step()
optimizer_s300.step()
#train_losses.append(loss.item())
train_losses_s300 = np.append(train_losses_s300, loss_s300.item())
train_losses_l300 = np.append(train_losses_l300, loss_l300.item())
#train_losses.append(loss.data)
#if epochs_done == 0:
#print(outputs_s300.shape)