def train_discriminator_step(self, configuration: Configuration, metadata: Metadata, architecture: Architecture,
batch: Batch) -> float:
# clean previous gradients
architecture.discriminator_optimizer.zero_grad()
# generate a batch of fake features with the same size as the real feature batch
fake_features = self.sample_fake(architecture, len(batch["features"]), condition=batch.get("labels"))
fake_features = fake_features.detach() # do not propagate to the generator
# calculate loss
loss = architecture.discriminator_loss(architecture,
batch["features"],
fake_features,
condition=batch.get("labels"))
# calculate gradients
loss.backward()
# update the discriminator weights
architecture.discriminator_optimizer.step()
# return the loss
return to_cpu_if_was_in_gpu(loss).item()
def train_discriminator_step(configuration: Configuration,
metadata: Metadata,
architecture: Architecture,
batch: Batch) -> float:
# clean previous gradients
architecture.discriminator_optimizer.zero_grad()
# generate a batch of fake features with the same size as the real feature batch
generated = architecture.generator(batch["features"],
missing_mask=batch["missing_mask"])
# replace the missing features by the generated
imputed = compose_with_mask(
mask=batch["missing_mask"],
differentiable=True, # now there are no NaNs and this should be used
where_one=generated,
where_zero=batch["raw_features"])
imputed = imputed.detach() # do not propagate to the generator
# generate hint
hint = generate_hint(batch["missing_mask"],
configuration.hint_probability, metadata)
# calculate loss
loss = architecture.discriminator_loss(
architecture=architecture,
imputed=imputed,
hint=hint,
missing_mask=batch["missing_mask"])
# calculate gradients
loss.backward()
# update the discriminator weights
architecture.discriminator_optimizer.step()
# return the loss
return to_cpu_if_was_in_gpu(loss).item()