def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
reconstruction_loss = self.create_other(
arguments.reconstruction_loss.factory, architecture, metadata,
arguments.reconstruction_loss.get("arguments", {}))
return AutoEncoderLoss(reconstruction_loss,
**arguments.get_all_defined(["masked"]))
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
reconstruction_loss = self.create_other(
arguments.reconstruction_loss.factory, architecture, metadata,
arguments.reconstruction_loss.get("arguments", {}))
optional_arguments = arguments.get_all_defined(
["reconstruction_loss_weight"])
return GAINGeneratorLoss(reconstruction_loss, **optional_arguments)
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
optional = arguments.get_all_defined(["sizes", "bn_decay", "shortcut"])
if "activation" in arguments:
activation_configuration = arguments.activation
optional["activation"] = self.create_other(
activation_configuration.factory, architecture, metadata,
activation_configuration.get("arguments", {}))
return HiddenLayersFactory(**optional)
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
# override the reduction argument
reconstruction_loss_configuration = arguments.reconstruction_loss.get(
"arguments", {})
if "reduction" in reconstruction_loss_configuration:
assert reconstruction_loss_configuration["reduction"] == "sum"
else:
reconstruction_loss_configuration["reduction"] = "sum"
# create the reconstruction loss
reconstruction_loss = self.create_other(
arguments.reconstruction_loss.factory, architecture, metadata,
reconstruction_loss_configuration)
# create the vae loss
return VAELoss(reconstruction_loss,
**arguments.get_all_defined(["masked"]))
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
return MultiInputDropout(
metadata, **arguments.get_all_defined(self.optional_arguments()))
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
return self.wrapped_class(
**arguments.get_all_defined(self.optional_class_arguments))
def impute(self, configuration: Configuration, metadata: Metadata,
scaled_inputs: Tensor, missing_mask: Tensor) -> Tensor:
optional = configuration.get_all_defined(["noise_mean", "noise_std"])
optional["differentiable"] = False
return NormalNoiseImputationLayer(**optional)(scaled_inputs,
missing_mask)
def create(self, architecture: Architecture, metadata: Metadata,
arguments: Configuration) -> Any:
return MeanAndModesImputationLayer(
to_gpu_if_available(
torch.from_numpy(np.load(arguments.path)).float()),
**arguments.get_all_defined(["differentiable"]))
def create(self, architecture: Architecture, metadata: Metadata, arguments: Configuration) -> Any:
return MultiReconstructionLoss(metadata, **arguments.get_all_defined(self.optional_arguments()))