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

[dask] Support more meta data. #6132

Merged
merged 5 commits into from
Sep 21, 2020
Merged

Conversation

trivialfis
Copy link
Member

  • label_lower_bound
  • label_upper_bound
  • base_margin

@trivialfis
Copy link
Member Author

@hcho3 I modified the test for survival training to use it for dask. Could you please help taking a look?

@codecov-commenter
Copy link

codecov-commenter commented Sep 19, 2020

Codecov Report

Merging #6132 into master will increase coverage by 0.16%.
The diff coverage is 94.23%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master    #6132      +/-   ##
==========================================
+ Coverage   78.67%   78.83%   +0.16%     
==========================================
  Files          12       12              
  Lines        3071     3090      +19     
==========================================
+ Hits         2416     2436      +20     
+ Misses        655      654       -1     
Impacted Files Coverage Δ
python-package/xgboost/core.py 78.34% <ø> (+0.26%) ⬆️
python-package/xgboost/dask.py 80.38% <94.23%> (+0.54%) ⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cc82ca1...6a49405. Read the comment docs.

Copy link
Collaborator

@hcho3 hcho3 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for adding survival analysis in the Dask test. Some comments:

python-package/xgboost/dask.py Show resolved Hide resolved
assert shap.ndim == 2
assert shap.shape[0] == kRows
assert shap.shape[1] == kCols + 1


def test_aft_survival():
# survival doesn't handle empty dataset well.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I did not know this. How can we fix this?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@hcho3 There are checks:

    CHECK_NE(info.labels_lower_bound_.Size(), 0U)
      << "labels_lower_bound cannot be empty";
    CHECK_NE(info.labels_upper_bound_.Size(), 0U)
      << "labels_upper_bound cannot be empty";

in survival_metric.cu::EvalEWiseSurvivalBase::Eval.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we remove the checks and have the objective return zero?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Depends on the synchronization I think.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

AFT is an element-wise objective, so it should behave like reg:squarederror when it comes to handling empty datasets. Let me create a new issue for this.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Opened #6139

@trivialfis trivialfis merged commit 33d80ff into dmlc:master Sep 21, 2020
@trivialfis trivialfis deleted the dask-meta-data branch September 21, 2020 08:56
@trivialfis trivialfis mentioned this pull request Sep 21, 2020
14 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants