emmi.functional.init

Attributes

ALL_CONVS
ALL_LAYERS

Functions

init_bias_to_zero(layer_module)	Initialize the bias tensor of a nn.Module instance to zero.
init_trunc_normal_zero_bias(layer_module[, std])	Initialize the weight tensor of a nn.Module instance using the truncated normal initialization with a zero bias vector.

Module Contents

emmi.functional.init.ALL_CONVS

emmi.functional.init.ALL_LAYERS

emmi.functional.init.init_bias_to_zero(layer_module)

Initialize the bias tensor of a nn.Module instance to zero.

Parameters:

layer_module (torch.nn.Module) – An nn.Module instance, either a Linear or Conv layer.

Return type:

None

emmi.functional.init.init_trunc_normal_zero_bias(layer_module, std=0.02)

Initialize the weight tensor of a nn.Module instance using the truncated normal initialization with a zero bias vector.

Parameters:

• layer_module (torch.nn.Module) – An nn.Module instance, either a Linear or Conv layer.

• std (float) – Standard Deviation value of the normal distribution to sample weights from. Defaults to 0.02.

Return type:

None