continual.forward_stepping¶

continual.forward_stepping(module, dim=2)[source]¶

Enhances torch.nn.Module with forward_step and forward_steps

Note

The passed module must not have time-dependent operations! For instance, module = nn.Conv3d(1, 1, kernel_size=(1,1,1)) is OK, but results for module = nn.Conv3d(1, 1, kernel_size=(3,3,3)) would be invalid.

Alternatively, one may attempt to automatically convert the module by using co.continual instead.

Parameters:

module (nn.Module) – the torch.nn.Module to enchance.
dim (int, optional) – The dimension to unsqueeze during forward_step. Defaults to 2.