Alpaca nn modules

class alpaca.nn.Module(*args, **kwargs)

The class links nn.Module with the alpaca Module abstraction by allowing us to copy nn.Module instance’s dictionary into this class instance. Additionally, the class introduces additional flags for the inference/uncertainty estimation modes.

instantiate_with_dropout_params(module: torch.nn.modules.module.Module, dropout_rate: float = 0.0, dropout_mask: BaseMask = None) → alpaca.nn.Module

Copies the instant nn.Module but also adding dropout_mask/dropout_rate parameters

Parameters

• module (nn.Module) – The instance nn.Module to be copied

• dropout_rate (float) – The dropout rate

• dropout_mask ("BaseMask") – Base mask instance setting the type of mask of the module

ue_mode() → alpaca.nn.Module

Sets the alpaca.Module into the uncertainty estimaton mode. This will enable the dropout mask logic calculation with the dropout rate activated.

inf_mode() → alpaca.nn.Module

Sets the alpaca.Module into inference mode. This will disable dropout_rate and dropout_mask of the module.

class alpaca.nn.Dropout(*args, dropout_rate: float = 0.0, dropout_mask: BaseMask = None, **kwargs)

The subclass of nn.Dropout layer with the additional dropout_mask and dropout_rate parameterization

Parameters

• dropout_rate (float) – Dropout rate of the mask

• dropout_mask ("BaseMask") – Base mask instance setting the type of mask of the module