continual.continual¶

continual.continual(module)[source]¶

Convert a torch.nn module to a Continual Inference Network enhanced with forward_step and forward_steps.

Modules may be either torch.nn Modules for which a corresponding module is implemented in this library (e.g. nn.Conv3d), or a torch.nn which can be naively mapped (e.g. nn.ReLU).

Custom modules can also be made continual by means of the forward_stepping function.

Examples

coconv = co.continual(nn.Conv3d(3, 3, 3))

coseq = co.continual(nn.Sequential(
    nn.Conv3d(3, 3, 3),  # Automatically converted as well!
    nn.ReLU()
))

Return type:: CoModule