-
Notifications
You must be signed in to change notification settings - Fork 54
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Implement sketched percentile #1420
Conversation
Thank you for the PR! |
Thank you for the PR! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1420 +/- ##
=======================================
Coverage 91.86% 91.86%
=======================================
Files 80 80
Lines 11860 11878 +18
=======================================
+ Hits 10895 10912 +17
- Misses 965 966 +1
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
Thank you for the PR! |
1 similar comment
Thank you for the PR! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Some minor comments on documentation
followed mtar's suggestion Co-authored-by: Michael Tarnawa <m.tarnawa@fz-juelich.de>
updated docstring according to review
for more information, see https://pre-commit.ci
Thank you for the PR! |
for more information, see https://pre-commit.ci
Thank you for the PR! |
1 similar comment
Thank you for the PR! |
updated docstring according to review
for more information, see https://pre-commit.ci
updated docstring according to review
for more information, see https://pre-commit.ci
Thank you for the PR! |
Thank you for the PR! |
Thank you for the PR! |
Thank you for the PR! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Just a couple tiny changes, otherwise I think we can merge this week. Thanks a lot @mrfh92 !
Co-authored-by: Claudia Comito <39374113+ClaudiaComito@users.noreply.github.com>
Co-authored-by: Claudia Comito <39374113+ClaudiaComito@users.noreply.github.com>
I think showing the example accuracies does not make so much sense since these accuracies very much depend on the data set (even if the median is quite stable w.r.t. outliers) and usually, we only show examples for small data that would allow calculation by hand, too. @ClaudiaComito except for that I have merged your two change requests. |
Thank you for the PR! |
1 similar comment
Thank you for the PR! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great, thanks @mrfh92 !
Thank you for the PR! |
…com:helmholtz-analytics/heat into features/1411-Implement_sketched_percentile
Thank you for the PR! |
@ClaudiaComito I have tried to fix the bug on torch 1.12 and 1.13 |
This solves #1411 by implementing a sketched version of percentile / median and a corresponding option for the RobustScaler in the preprocessing module.
All new features integrate into the current functionality without changes, due to choice of the current non-sketched version as default.
Performance comparison
on 14 MPI-processes (Workstation, CPU) on behalf of the "asteroids data set" (1317275 data points, 9 features), times and deviation measured over10 runs
*relative deviation of the sketched median vs true median per feature, averaged over 10 runs; only relative deviation for the feature with maximal deviation is reported