Skip to content
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

Avoid thread block with sparse data. #7255

Merged
merged 6 commits into from Sep 25, 2021

Conversation

trivialfis
Copy link
Member

The 0.5 threshold is just a heuristic not a result of a rigorous benchmark. For now, I need to close #6659 since sparse datasets are quite usual with one-hot encoding.

@ShvetsKS

@ShvetsKS
Copy link
Contributor

ShvetsKS commented Sep 24, 2021

@trivialfis LGTM in general. But is it possible in future to have the possibility to disable "memory saving mode" for sparse case (if performance is dropped)?

@trivialfis
Copy link
Member Author

@ShvetsKS Thanks for the quick response. In general, I'm open to setting the threshold to smaller values when needed or simply use binary search for extremely sparse datasets. I think we can find a good threshold for handling those datasets. Also now that we have a new prediction API that accepts JSON input, you can add a new parameter if needed as the last option.

@trivialfis trivialfis merged commit d8a549e into dmlc:master Sep 25, 2021
@trivialfis trivialfis deleted the avoid-sparse-thread-block branch September 25, 2021 05:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Memory usage jumps to 50G when trying to predict
2 participants