allennlp.modules.residual_with_layer_dropout¶

class allennlp.modules.residual_with_layer_dropout.ResidualWithLayerDropout(undecayed_dropout_prob: float = 0.5)[source]

Bases: torch.nn.modules.module.Module

A residual connection with the layer dropout technique Deep Networks with Stochastic Depth .

This module accepts the input and output of a layer, decides whether this layer should be stochastically dropped, returns either the input or output + input. During testing, it will re-calibrate the outputs of this layer by the expected number of times it participates in training.

forward(layer_input: torch.Tensor, layer_output: torch.Tensor, layer_index: int = None, total_layers: int = None) → torch.Tensor[source]

Apply dropout to this layer, for this whole mini-batch. dropout_prob = layer_index / total_layers * undecayed_dropout_prob if layer_idx and total_layers is specified, else it will use the undecayed_dropout_prob directly.

Parameters: layer_input torch.FloatTensor required The input tensor of this layer. layer_output torch.FloatTensor required The output tensor of this layer, with the same shape as the layer_input. layer_index int The layer index, starting from 1. This is used to calcuate the dropout prob together with the total_layers parameter. total_layers int The total number of layers. output: torch.FloatTensor A tensor with the same shape as layer_input and layer_output.