issues
search
accosmin-org
/
libnano
C++ numerical optimization and machine learning utilities using Eigen3
MIT License
4
stars
1
forks
source link
[gboost] extend stump and hinge weak learners to use quantiles
#53
Open
accosmin
opened
2 years ago
accosmin
commented
2 years ago
the threshold can be fixed to a given range of quantiles (e.g. from 5% to 95% in 5% increments)
it should increase robustness to outliers and improve generalization
extend the unit tests accordingly