-
Notifications
You must be signed in to change notification settings - Fork 3.4k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[Metrics] class based embedding similarity + tests (#3358)
* embedding similarity class + test * fix tests * fix pep8 * add docs * noindex * Update docs/source/metrics.rst * Update pytorch_lightning/metrics/self_supervised.py Co-authored-by: Rohit Gupta <[email protected]> * Update pytorch_lightning/metrics/self_supervised.py Co-authored-by: Rohit Gupta <[email protected]> * suggestions * changes to init * move __all__ * fix imports * Apply suggestions from code review * assert typo * change import Co-authored-by: Adrian Wälchli <[email protected]> Co-authored-by: Justus Schock <[email protected]> Co-authored-by: Rohit Gupta <[email protected]> Co-authored-by: Jirka Borovec <[email protected]> Co-authored-by: Nicki Skafte <[email protected]>
- Loading branch information
6 people
authored
Sep 11, 2020
1 parent
70af47d
commit 93cf6d0
Showing
6 changed files
with
130 additions
and
9 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,73 @@ | ||
| # Copyright The PyTorch Lightning team. | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| from typing import Any | ||
|
|
||
| import torch | ||
|
|
||
| from pytorch_lightning.metrics.functional.self_supervised import embedding_similarity | ||
| from pytorch_lightning.metrics.metric import TensorMetric | ||
|
|
||
|
|
||
| class EmbeddingSimilarity(TensorMetric): | ||
| """ | ||
| Computes similarity between embeddings | ||
| Example: | ||
| >>> embeddings = torch.tensor([[1., 2., 3., 4.], [1., 2., 3., 4.], [4., 5., 6., 7.]]) | ||
| >>> embedding_similarity(embeddings) | ||
| tensor([[0.0000, 1.0000, 0.9759], | ||
| [1.0000, 0.0000, 0.9759], | ||
| [0.9759, 0.9759, 0.0000]]) | ||
| """ | ||
| def __init__( | ||
| self, | ||
| similarity: str = 'cosine', | ||
| zero_diagonal: bool = True, | ||
| reduction: str = 'mean', | ||
| reduce_group: Any = None | ||
| ): | ||
| """ | ||
| Args: | ||
| similarity: 'dot' or 'cosine' | ||
| reduction: 'none', 'sum', 'mean' (all along dim -1) | ||
| zero_diagonal: if True, the diagonals are set to zero | ||
| reduce_group: the process group to reduce metric results from DDP | ||
| """ | ||
| super().__init__(name='embedding_similarity', | ||
| reduce_group=reduce_group) | ||
| assert similarity in ('dot', 'cosine') | ||
| self.similarity = similarity | ||
| isinstance(zero_diagonal, bool) | ||
| self.zero_diagonal = zero_diagonal | ||
| assert reduction in ('none', 'sum', 'mean') | ||
| self.reduction = reduction | ||
|
|
||
| def forward(self, batch: torch.Tensor) -> torch.Tensor: | ||
| """ | ||
| Actual metric computation | ||
| Args: | ||
| batch: tensor containing embeddings with shape (batch_size, dim) | ||
| Return: | ||
| A square matrix (batch, batch) with the similarity scores between all elements | ||
| If sum or mean are used, then returns (b, 1) with the reduced value for each row | ||
| """ | ||
| return embedding_similarity(batch, | ||
| similarity=self.similarity, | ||
| zero_diagonal=self.zero_diagonal, | ||
| reduction=self.reduction) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,35 @@ | ||
| import pytest | ||
| import torch | ||
| from sklearn.metrics import pairwise | ||
|
|
||
| from pytorch_lightning.metrics.functional.self_supervised import embedding_similarity | ||
|
|
||
|
|
||
| @pytest.mark.parametrize('similarity', ['cosine', 'dot']) | ||
| @pytest.mark.parametrize('reduction', ['none', 'mean', 'sum']) | ||
| def test_against_sklearn(similarity, reduction): | ||
| """Compare PL metrics to sklearn version.""" | ||
| device = 'cuda' if torch.cuda.is_available() else 'cpu' | ||
|
|
||
| batch = torch.randn(5, 10, device=device) # 100 samples in 10 dimensions | ||
|
|
||
| pl_dist = embedding_similarity(batch, similarity=similarity, | ||
| reduction=reduction, zero_diagonal=False) | ||
|
|
||
| def sklearn_embedding_distance(batch, similarity, reduction): | ||
|
|
||
| metric_func = {'cosine': pairwise.cosine_similarity, | ||
| 'dot': pairwise.linear_kernel}[similarity] | ||
|
|
||
| dist = metric_func(batch, batch) | ||
| if reduction == 'mean': | ||
| return dist.mean(axis=-1) | ||
| if reduction == 'sum': | ||
| return dist.sum(axis=-1) | ||
| return dist | ||
|
|
||
| sk_dist = sklearn_embedding_distance(batch.cpu().detach().numpy(), | ||
| similarity=similarity, reduction=reduction) | ||
| sk_dist = torch.tensor(sk_dist, dtype=torch.float, device=device) | ||
|
|
||
| assert torch.allclose(sk_dist, pl_dist) |