Closed mfasiolo closed 8 years ago
TuneLearn and TuneLearnFast ended up using the lambda (which is a vector) used by the main fit, without taking into account that they are using different data when bootstrapping! This results in a total mess when the heteroscedasticity is strong.
Hopefully solved here:
https://github.com/mfasiolo/qgam/commit/50a0cabc6c087bc64d9600e82bdbbeb3a5e17845
TuneLearn and TuneLearnFast ended up using the lambda (which is a vector) used by the main fit, without taking into account that they are using different data when bootstrapping! This results in a total mess when the heteroscedasticity is strong.