def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
sample = get_tracer(idx)
if self.transform:
sample = self.transform(sample)
return sample
errG.backward()
optimizerG.step()
#Display losses
if i_batch % 50 == 0:
print("Training Losses")
print("Epoch: ", epoch, " | i: ", i_batch)
print("Discriminator Loss: ", errD.item())
print("Generator Loss: ", errG.item())
print("Gen BCE Loss:", errBCE.item())
print("Gen MSE Loss:", alpha * errMSE.item())
## Pass through networks and calculate losses
errG_val_mse = 0
for i in val_ints:
val_tracer = get_tracer(i)
val_tracer = torch.from_numpy(val_tracer).unsqueeze(1).to(
device=device, dtype=torch.float)
val_tracer_incr = get_tracer(i + 1)
val_tracer_incr = torch.from_numpy(val_tracer_incr).unsqueeze(1).to(
device=device, dtype=torch.float)
# Pass through Encoder + Generator
val_outputEnc = netEnc(val_tracer).detach()
val_outputG = netG(val_outputEnc).detach()
errG_val_mse += mse_loss(val_outputG, val_tracer_incr)
## Display Validation losses
errG_val_mse /= len(val_ints)
errG_val_mse *= alpha
errAE = mse_loss(output, data)
errAE.backward()
optimizerEnc.step()
optimizerDec.step()
if i_batch % 50 == 0:
print("Test Loss:")
print("Epoch: ", epoch, " | i: ", i_batch)
print("AutoEncoder Loss: ", errAE.item())
# Get error for Validation set
## Pass through networks and calculate losses
errAE_Val = 0
for i in val_ints:
val_tracer = get_tracer(i)
val_tracer = torch.from_numpy(val_tracer).unsqueeze(1).to(
device=device, dtype=torch.float)
# Pass through Encoder
val_output = netEnc(val_tracer).detach()
val_output = netDec(val_output).detach()
errAE_Val += mse_loss(val_output, val_tracer)
errAE_Val /= len(val_ints)
# Storing losses per epoch to plot
epoch_list.append(epoch)
loss_list.append(errAE.item())
val_loss_list.append(errAE_Val.item())
torch.cuda.is_available() and ngpu > 0) else "cpu")
#################
# Instantiating #
#################
netEnc = Encoder(ngpu).to(device)
netG = Generator(ngpu).to(device)
checkpoint = torch.load(filePathToModelAE)
netEnc.load_state_dict(checkpoint['netEnc_state_dict'])
checkpoint = torch.load(filePathToModelGAN)
netG.load_state_dict(checkpoint['netG_state_dict'])
mse_loss = nn.MSELoss()
tracer_dataset = TracerDataset(transform=ToTensor())
batch_indicies = []
for i in range(3729):
batch_indicies.append(i)
for i in batch_indicies:
data = get_tracer(i)
data = torch.from_numpy(data).unsqueeze(0).to(device=device,
dtype=torch.float)
output = denormalise(netG(netEnc(data)), x_min, x_max)
output = np.array(output.squeeze().cpu().detach())
create_tracer_VTU_GAN(i, output, "tGAN")
print(i)