nagataka / pytorch-code-reading Goto Github PK

Simple and familiar thing first...

pytorch/torch/nn/modules/loss.py (class: MSELoss)

Simply call F.mse_loss(input, target, reduction=self.reduction) and return it.

pytorch/torch/nn/functional.py(function: mse_loss)

First, compute squared error ret = (input - target) ** 2.
Then, perform reduction based on parameter: 'mean' or 'sum' (default: 'mean') ret = torch.mean(ret) if reduction == 'mean' else torch.sum(ret) where reduction is defined in pytorch/torch/nn/_reduction.py which simply returns 0(none), 1(mean), or 2(sum)