Reduce memory usage in gKDR #244
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
- Possible since the regularized Gram matrix is Hermitian - Use some more descriptive variable names
edaub
approved these changes
Sep 7, 2022
|
@edaub - thanks for the review - I've now finally resolved your comments above. I've left the target of the PR as main but the version number for you to bump up as discussed. |
|
I've decided to stick the three current outstanding bugfixes into a single bugfix release 0.7.2, so changing the base branch to put them all onto a single temporary branch prior to merging. I'll also merge this into |
edaub
added a commit
that referenced
this pull request
Nov 4, 2022
* added bugfix to delete GPU .so file committed to devel to bugfix release on main * Update documentation to reflect patsy requirement (#245) A previous major upgrade added patsy as a requirement rather than an optional dependency. The documentation (readme and installation page on the sphinx docs) now have been updated to reflect this. * Reduce memory usage in gKDR (#244) * Test that a larger gKDR example can run * Reduce memory footprint in gKDR with an explicit loop * Use cholesky/cho_solve instead of general linear solver - Possible since the regularized Gram matrix is Hermitian - Use some more descriptive variable names * Add explanatory docstring * Generate deterministic test data for 'large' gKDR test * Additional docstrings * Use cho_factor instead of cholesky, which simplifies immediate calls to cho_solve Co-authored-by: ots22 <ots22@users.noreply.github.com>
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Outdated suggestions cannot be applied.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.
Fixes #242.
The current implementation of gKDR exclusively uses array computations. A drawback of this is that there are some large temporary arrays created (the largest having size
N^2 M^2whereNis the number of observations andMis the number of features).This PR replaces one of these array computations with an explicit loop, reducing the space used to
O(M^2 + M N^2).It also replaces two calls to
np.linalg.solvewith calls tocho_solve(and a single Cholesky decomposition).