ATOMScience-org / AMPL

The ATOM Modeling PipeLine (AMPL) is an open-source, modular, extensible software pipeline for building and sharing models to advance in silico drug discovery.
MIT License
136 stars 68 forks source link

Handle the case when split subsets is empty #343

Closed mauvais2 closed 2 months ago

mauvais2 commented 3 months ago

From 1.6.2 testing, found these errors:

=================================== FAILURES ===================== FAILED test_split.py::test_splits - assert 0.31782106782106784 <= 0.313725490... FAILED test_split.py::test_pipeline_split_only - ValueError: Distribution can... FAILED test_split.py::test_pipeline_split_and_train - ValueError: Distributio... ================== 3 failed, 12 warnings in 962.65s (0:16:02) ==================

Kevin said the failure is in the code that computes the Wasserstein distance between response value distributions in the split subsets where it is empty. A workaround will be provided.

mauvais2 commented 2 months ago

Fixed in https://github.com/ATOMScience-org/AMPL/commit/a001100decbfb50cf09de49aacb514d06a4471ce