neulab / ExplainaBoard

Interpretable Evaluation for AI Systems
MIT License
360 stars 36 forks source link

Add seed argument to Metric #565

Closed tetsuok closed 1 year ago

tetsuok commented 1 year ago

Overview

This PR adds seed argument to Metric class to fix #537.

Details

Metric's calc_confidence_interval method uses random generator for bootstrapping, but we have no way to provide a seed to the method to make the behavior reproducible. To address this issue, adds seed argument to Metric class, as described in https://github.com/neulab/ExplainaBoard/issues/537#issuecomment-1274340748.

As per this change, reverted a hot fix in #550. This is to use almost equal assertions to assert confidence intervals computed by bootstrapping.

References

Blocked by