def _test_g(self, G: ModelTrainer, real):
gen_pred = torch.nn.functional.sigmoid(G.forward(real))
G.compute_and_update_test_loss("MSELoss", gen_pred, real)
metric = G.compute_metric("MeanSquaredError", gen_pred, real)
G.update_test_metric("MeanSquaredError", metric / 32768)
return gen_pred
def _train_g(self, G: ModelTrainer, real, backward=True):
G.zero_grad()
gen_pred = torch.nn.functional.sigmoid(G.forward(real))
loss_G = G.compute_and_update_train_loss("MSELoss", gen_pred, real)
metric = G.compute_metric("MeanSquaredError", gen_pred, real)
G.update_train_metric("MeanSquaredError", metric / 32768)
if backward:
loss_G.backward()
G.step()
return gen_pred