| from typing import Optional, Dict |
|
|
| import mxnet |
|
|
| from . import RearrangeMixin, ReduceMixin |
| from ._einmix import _EinmixMixin |
|
|
| __author__ = 'Alex Rogozhnikov' |
|
|
|
|
| class Rearrange(RearrangeMixin, mxnet.gluon.HybridBlock): |
| def hybrid_forward(self, F, x): |
| return self._apply_recipe(x) |
|
|
|
|
| class Reduce(ReduceMixin, mxnet.gluon.HybridBlock): |
| def hybrid_forward(self, F, x): |
| return self._apply_recipe(x) |
|
|
|
|
| class EinMix(_EinmixMixin, mxnet.gluon.HybridBlock): |
| def _create_parameters(self, weight_shape, weight_bound, bias_shape, bias_bound): |
| with self.name_scope(): |
|
|
| self.weight = self.params.get(name='weight', shape=weight_shape, |
| init=mxnet.initializer.Uniform(weight_bound), |
| ) |
| if bias_shape is not None: |
| self.bias = self.params.get(name='bias', shape=bias_shape, |
| init=mxnet.initializer.Uniform(bias_bound), |
| ) |
| else: |
| self.bias = None |
|
|
| def _create_rearrange_layers(self, |
| pre_reshape_pattern: Optional[str], |
| pre_reshape_lengths: Optional[Dict], |
| post_reshape_pattern: Optional[str], |
| post_reshape_lengths: Optional[Dict]): |
| if (pre_reshape_pattern is not None) or (post_reshape_pattern is not None): |
| raise NotImplementedError("EinMix in mxnet/gluon doesn't support axis group/ungroup " |
| "because einsum in gluon defined only for mx.np.ndarrays") |
|
|
| def hybrid_forward(self, F, x, *args, **kwargs): |
| |
| |
| result = mxnet.np.einsum(self.einsum_pattern, x.as_np_ndarray(), self.weight.data()) |
| if self.bias is not None: |
| result += self.bias.data() |
| return result |
|
|
