Hi,

I'd like to propose a couple of features that, as a user, I believe may be
useful for Gaussian mixture modeling with GaussianMixture. I don't see how
to currently do them, and I think they should be quite easy to implement,
but I may be wrong.

   1. The weights of the components may be known. In that case, we would
   like to provide them through weights_init, but then they should be frozen
   and not optimized by the EM steps.
   So perhaps a frozen_weights boolean parameter, relevant only when
   weights_init is not None?
   2. In covariance_type, there are the full/diag/spherical options, but I
   think that logically "tied" is independent - currently it's "tied full",
   but there could just as well be "tied diag" and even "tied spherical" with
   only a single parameter for the variances. So I propose having two input
   parameters allowing for 3*2 = 6 structures instead of the current 4.

Thanks for all your great work for the community,
Ehud Schreiber.
_______________________________________________
scikit-learn mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3//lists/scikit-learn.python.org
Member address: [email protected]

Reply via email to