You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There is a missespecification on _get_kernel method at KernelPCA class.
If you perform the inverse_transform method specifiying gamma as None or directly imputing the value 1 / X.shape[1], it will differ, as the _get_kernel function with gamma as None adjusts to the X data both in transform and at inverse_transform methods.
The consequence is that specifying the supposed default hyperparameter leads to different results than leaving it as None.
The proposal is to add a hyperparameter called inverse_gamma or something like that for disambiguation.
The text was updated successfully, but these errors were encountered:
Guillermogsjc
changed the title
_get_kernel missespecified on KernelPCA
gamma hypeparameter missespecified on KernelPCA
Feb 6, 2019
Guillermogsjc
changed the title
gamma hypeparameter missespecified on KernelPCA
gamma hyperparameter misspecified on KernelPCA
Feb 6, 2019
mmm, have to check the other kernel hyperparams that should have an "inverse" option I guess... coef0 and degree. As soon as I have time I will check it.
There is a missespecification on
_get_kernel
method atKernelPCA
class.If you perform the
inverse_transform
method specifiyinggamma
asNone
or directly imputing the value1 / X.shape[1]
, it will differ, as the_get_kernel
function withgamma
asNone
adjusts to theX
data both intransform
and atinverse_transform
methods.The consequence is that specifying the supposed default hyperparameter leads to different results than leaving it as
None
.The proposal is to add a hyperparameter called
inverse_gamma
or something like that for disambiguation.The text was updated successfully, but these errors were encountered: