Add loss: Noise Constrastive Estimation #65
Conversation
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
christophmluscher
requested review from
albertz,
NeoLegends,
JackTemaki,
Gerstenberger,
michelwi and
Atticus1806
albertz
reviewed
Dec 19, 2024
Co-authored-by: Albert Zeyer <[email protected]>
curufinwe
requested changes
Jan 10, 2025
|
Added some small comments about documentation. Otherwise looks fine to me. |
JackTemaki
requested changes
Jan 16, 2025
i6_models/parts/losses/nce.py
Outdated
| def forward(self, data: torch.Tensor, target: torch.Tensor) -> torch.Tensor: | ||
| """ | ||
|
|
||
| :param data: the tensor for the input data, where batch and time are flattened resulting in [B x T, F] shape. |
There was a problem hiding this comment.
The comments here are more confusing than helpful. So it should be specified what data is (or even rename it), First I assumed it is logits, but then I guess this should be the input to the "output" layer of the provided model? This is very confusing and unclear. Then it should also be clear what is the input shape. When you say "where batch and time are flattened resulting in" this sounds like this is done internally, but what you mean it expects [B x T, F] as only valid input, correct?
curufinwe
requested changes
Jan 17, 2025
| :param model: model on which the NCE loss is to be applied. The model requires a member called `output`, which | ||
| is expected to be a linear layer with bias and weight members. The member `output` represents the output | ||
| layer of the model allowing access to the parameters during loss computation. | ||
| :param noise_distribution_sampler: for example `i6_model.samplers.LogUniformSampler`. |
There was a problem hiding this comment.
Suggested change
| :param noise_distribution_sampler: for example `i6_model.samplers.LogUniformSampler`. | |
| :param noise_distribution_sampler: for example `i6_models.parts.samplers.LogUniformSampler`. |
|
|
||
| :param num_samples: num of samples for the estimation, normally a value between 1000-4000. | ||
| 2000 is a good starting point. | ||
| :param model: model on which the NCE loss is to be applied. The model requires a member called `output`, which |
There was a problem hiding this comment.
Why not pass in these two weights explicitly?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add NCE loss
implemented by @Gerstenberger
depends on #64