L1-regularization for a single matrix

[rasoulnorouzi]

See This:

How to Implement Custom Regularization Losses on the Weights?

This what I do: class YourModel(nn.Module): def __init__(self,x,y): super(MF, self).__init__() pass def init_weights(self): pass def regularization(self): return your regularization term # example: return torch.sum(torch.pow(self.x.weight, 2)) + torch.sum(torch.pow(self.y.weight, 2)) def forward(self): pass and for Loss: loss = criterion(pred, target) + lambda*(model.regularization())