search

MarconiS / TraitsOnHeaven

0 stars 0 forks source link

What's up with GP #5

Open sjgraves opened 7 years ago

sjgraves commented 7 years ago

Sergio will look into what's going on with GP

MarconiS commented 6 years ago

update at 09/12: GP, SVMR, PLS working as an ensemble of the same model trained on the same 1000 permutations of pixels. Results range around 0.2-0.6 for most of the traits. Interestingly, GP and SVMR fail in getting the relationship for different traits.
Both GP and SVMR are based on a gaussian kernel, sklearn standard initial parameterization.