RosettaCommons / RosettaCommons/RFdiffusion

different protein samples

Open
#397 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
3.1k
Forks
644
PR merge metrics
No merged PRs in 30d

Description

Dear Developers, thank you for your help!

I would like to ask: how does RF handle the problem of excessively large loss deviations across different timesteps t and different protein samples during training?
When I use rigid-body translation MSE loss, I find that the loss fluctuates a lot — ranging from thousands down to less than 1. The overall average loss also seems to oscillate, because it may be stuck in a shaking stage, almost without effective learning. Even when I print the average loss at t < 0.2, I find that it still does not show effective learning.

I look forward to your guidance, thank you! Wish you a pleasant day!

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No files, tests, or entry points are named. Start by locating the RF training loss calculation and tracing how rigid-body translation MSE is aggregated across timesteps and protein samples. Compare the per-sample and per-timestep loss distributions, with attention to t < 0.2; done means explaining whether the fluctuation is expected or identifying a concrete training issue.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.