dotnet / dotnet/machinelearning
Models produced by the GAM Trainer depend on feature flocks
Open
area-Trees
bug
Priority:1
- Dominant language
- C#
- Stars
- 9.4k
- Forks
- 2k
- Avg merge
- 2d 20h
- Merged PRs (30d)
- 11
Description
In `FastTree`, `Feature Flocks` are simply helpful tools in the calculation and do not impact the final model. When `GAMs` are trained however, the final model is dependent on the `Feature Flocks`. That is, if there are 6 features, and two are flocked, then the resulting model will have 5 shape functions and one of the input features will be ignored.
As per `FastTree`, the output of `GAM` training should not depend on whether feature flocks were used in the calculation.
Contributor guide
Assessment
This issue has not been assessed yet.