triton-inference-server/fil_backend

[BUG] cuml binary classification models do not observe threshold

RAMitchell opened this issue · 1 comments

When setting threshold=0.9, cuml models for CPU or GPU do not appear to observe the threshold.

Other model types seem to be working.

This is occurring because these binary classification models are stored by treelite as vector leaf models, which are handled differently in postprocessing code and do not account for thresholds.