Module init

Variable initialization.

Enums

FanInOut

Number of features as input or output of a layer. In Kaiming initialization, choosing FanIn preserves the magnitude of the variance of the weights in the forward pass, choosing FanOut preserves this magnitude in the backward pass.

Init

Variable initializations.

NonLinearity

The non-linear function that follows this layer. ReLU is the recommended value.

NormalOrUniform

Constants

DEFAULT_KAIMING_NORMAL

DEFAULT_KAIMING_UNIFORM

Functions

f_init

Creates a new float tensor with the specified shape, device, and initialization.

init

Creates a new float tensor with the specified shape, device, and initialization.