(self, x, dropout=0.5)
| 76 | super(LockedDropout, self).__init__() |
| 77 | |
| 78 | def forward(self, x, dropout=0.5): |
| 79 | if not self.training or not dropout: |
| 80 | return x |
| 81 | m = x.data.new(1, x.size(1), x.size(2)).bernoulli_(1 - dropout) |
| 82 | mask = Variable(m.div_(1 - dropout), requires_grad=False) |
| 83 | mask = mask.expand_as(x) |
| 84 | return mask * x |
| 85 | |
| 86 | |
| 87 | def mask2d(B, D, keep_prob, cuda=True): |