[matnormal] Matrix-normal (dual probabilistic) SRM #490
Conversation
This file contains hidden or 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
Update with mainline repo
…sman/brainiak into matnormal-regression-rsa
Fixing style issues in utils.py
…iniak into matnormal-regression-rsa
…olve this automatically)
Member
|
Thank you, @mshvartsman! Could you please name a preferred owner of that second pair of eyes? |
Contributor
Author
Contributor
|
I would encourage others to review, as I am still owing review for another PR. |
Member
|
@szorowi1, could you please review this pull request? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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.
This is a TF2 port of the matrix-normal SRM code we used in https://arxiv.org/abs/1711.03058. It's not quite ready for merge but I'm putting up the PR to start discussion and maybe get another pair of eyes on the code. In contrast to the code we used for the paper (https://github.com/mshvartsman/brainiak/blob/feature-dpmnsrm/brainiak/matnormal/dpmnsrm.py), this is runnable in current TF and includes asserts to ensure the monotonicity of the ECM(E) algorithm. I also fixed at least one math bug (there's a typo in the paper that made it into the code). For now I'm excluding all model variants that require pymanopt (until pymanopt/pymanopt#126 is merged).
The bad news: those asserts fail on the update for subject-specific noise scalers (
rhoprec) and I can't figure out why. I'd love another pair of eyes on this so we can get it merged.Important disclaimer: This work is all done on my own time and my own equipment, based on a paper I wrote while at Princeton, and is unrelated to my current work with my employer.