Deep clustering/Chimera recipe #96
Merged
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
mpariente
commented
May 11, 2020
Comment on lines
+65
to
+67
| proj = proj.view(batch, n_frames, -1, self.embedding_dim).transpose(1, 2) | ||
| # (batch, freq * frames, emb) | ||
| proj = proj.reshape(batch, -1, self.embedding_dim) |
There was a problem hiding this comment.
The bug was here. Without the transpose, the time bins where not aligned with each other and training was impossible.
I added a note about it in the DC loss.
mpariente
commented
May 11, 2020
Comment on lines
+146
to
+155
| try: | ||
| # Last best model summary | ||
| with open(os.path.join(exp_dir, 'best_k_models.json'), "r") as f: | ||
| best_k = json.load(f) | ||
| best_model_path = min(best_k, key=best_k.get) | ||
| except FileNotFoundError: | ||
| # Get last checkpoint | ||
| all_ckpt = os.listdir(os.path.join(exp_dir, 'checkpoints/')) | ||
| all_ckpt.sort() | ||
| best_model_path = os.path.join(exp_dir, 'checkpoints', all_ckpt[-1]) |
There was a problem hiding this comment.
Also, this would be a way to bypass the best_k_models.json.
mpariente
commented
May 11, 2020
| with open(os.path.join(exp_dir, "best_k_models.json"), "w") as f: | ||
| json.dump(checkpoint.best_k_models, f, indent=0) | ||
| #torch.save(system.model.state_dict(), os.path.join(exp_dir, 'final.pth')) | ||
| # Save last model for convenience | ||
| torch.save(system.model.state_dict(), |
There was a problem hiding this comment.
And this would be another one. But if training didn't finish, this model is not saved..
Closed
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Outdated suggestions cannot be applied.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.
Adding the Deep clustering / Chimera++ recipe on wsj2mix and wsj3mix.
Mainly based on @sunits initial work.
Data prep, dataloader, training and evaluation script.
Evaluatio script will use the mask-inference head whenever possible and DC head only if loss_alpha is equal to 1.
First successful iteration on the DC head only gets 10.1 dB SDR improvement (not SI-SDR).
Things left to do :