duo_ai.utils.model

Functions

orthogonal_init(module[, gain])	Apply orthogonal initialization to a module's weights and zero to biases.
xavier_uniform_init(module[, gain])	Apply Xavier uniform initialization to a module's weights and zero to biases.

Module Contents

duo_ai.utils.model.orthogonal_init(module, gain=nn.init.calculate_gain('relu'))

Apply orthogonal initialization to a module’s weights and zero to biases.

Parameters:

• module (nn.Module) – The module to initialize (e.g., nn.Linear, nn.Conv2d).

• gain (float, optional) – Scaling factor for the orthogonal initialization. Default is relu gain.

Returns:

The initialized module.

Return type:

nn.Module

Examples

>>> layer = nn.Linear(10, 5)
>>> orthogonal_init(layer)

duo_ai.utils.model.xavier_uniform_init(module, gain=1.0)

Apply Xavier uniform initialization to a module’s weights and zero to biases.

Parameters:

• module (nn.Module) – The module to initialize (e.g., nn.Linear, nn.Conv2d).

• gain (float, optional) – Scaling factor for the Xavier initialization. Default is 1.0.

Returns:

The initialized module.

Return type:

nn.Module

Examples

>>> layer = nn.Linear(10, 5)
>>> xavier_uniform_init(layer)