-
Notifications
You must be signed in to change notification settings - Fork 53
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Speed-up and fix bug in condition evaluation during training.
The training condition evaluation is composed of a loop over the examples and a switch over the condition type (and a few other things). Prior to this change, the example loop was outside the condition switch loop, forcing the algorithm to re-check the condition type (and other things) for each examples. After this change, the condition type is outside of the example loop. Example of speed-ups: 1. Average speed-up of 3.7% on all benchmark. 2. Speed-up of 10-15% on Adult dataset with GBT. 3. Speed-up of 8% on Adult dataset with RF. 4. Speed-up of 9% on 4M dataset with 200 features with discretized GBT. 5. No speed difference (<1% gain) on 4M dataset with 200 features non-discretized GBT. Note: The absolute gain is the same as 4., but since 5.'s training is longer, the relative gain is insignificant. PiperOrigin-RevId: 670888905
- Loading branch information
1 parent
abe57f6
commit c407154
Showing
8 changed files
with
411 additions
and
24 deletions.
There are no files selected for viewing
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
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
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
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
Oops, something went wrong.