Loss improvements

[RoniGurvich]

@CodiumAI-Agent /review

[CodiumAI-Agent]

PR Analysis

• 🎯 Main theme: This PR primarily focuses on improving the loss function used in the project. It includes changes to the margin, scaling, and dimensions of the embeddings.

• 📌 Type of PR: Enhancement

• 🧪 Relevant tests added: True

• ✨ Focused PR: Yes, all the changes are related to the improvement of the loss function.

• 🔒 Security concerns: No security concerns found

  PR Feedback

• General suggestions: The PR seems to be well-structured and focused on a specific task, which is improving the loss function. The changes made to the margin, scaling, and dimensions of the embeddings seem to be reasonable. However, it would be beneficial to provide a description for the PR to give more context about the changes and their expected impact.

• 🤖 Code feedback:

  • relevant file: peptriever/finetuning/loss.py suggestion: It would be beneficial to add some comments to the code to explain the logic behind the changes, especially the calculation of the loss. This will make the code easier to understand for other developers. [medium] relevant line: loss = loss.filldiagonal(0).sum(dim=-1) / (loss.shape[1] - 1)

  • relevant file: peptriever/finetuning/config.py suggestion: The change in the 'encoded_sequence_dims' parameter from 128 to 512 is quite significant. It would be helpful to provide some explanation or justification for this change in the comments. [medium] relevant line: encoded_sequence_dims: int = 512

  • relevant file: peptriever/test/test_loss.py suggestion: The tests seem to be well-written and cover the main functionality of the loss function. However, it would be beneficial to add some edge cases to ensure the robustness of the function. [medium] relevant line: loss = loss_f({"y1": peps, "y2": prots}, None)

How to use

Tag me in a comment '@CodiumAI-Agent' and add one of the following commands: /review [-i]: Request a review of your Pull Request. For an incremental review, which only considers changes since the last review, include the '-i' option. /describe: Modify the PR title and description based on the contents of the PR. /improve: Suggest improvements to the code in the PR. /ask \<QUESTION>: Pose a question about the PR.

To edit any configuration parameter from 'configuration.toml', add --config_path=new_value For example: /review --pr_reviewer.extra_instructions="focus on the file: ..." To list the possible configuration parameters, use the /config command.

[RoniGurvich]

Closing to continue in a fork repo until the next version is ready to release