Fix deviance calculations

andland / generalizedPCA

Dimensionality reduction for exponential family data by extending PCA

Other

23 stars 2 forks source link

Closed andland closed 8 years ago

andland commented 9 years ago

[x] Percent of deviance explained is incorrect for Poisson, I think. I believe this has to do with not subtracting the saturated model
[x] Also, the null deviance is calculated incorrectly when there are weights
[ ] May also want to modify convergence criteria to percent change in loss. Need to determine what a good percent cutoff is.

andland commented 8 years ago

The last sub-issue is unrelated to deviance calculations, so closing and opening a new issue