GaussianMixture is runnning too many computation when parameters are alredy provided

[glemaitre]

In GaussianMixture, if a user is passing some initial values for the weights, means, and precision then there is no need to run the initialization (via kmeans or random) and to estimate the gaussian parameters.

However, currently, we still run these two steps and then discard them.

I think that we can make the estimator more efficient by bypassing these steps in case we already have all the necessary statistics to start the EM algorithm.

[jiawei-zhang-a]

If it's acceptable, may I kindly take responsibility for addressing this issue?

[glemaitre]

@jiawei-zhang-a go ahead.

[jiawei-zhang-a]

@glemaitre Thank you so much!

[jiawei-zhang-a]

/take

[jiawei-zhang-a]

@glemaitre Sir, I have a question about adding private variable into class and where to add it. I do not know if I can add a private variable _init_skipped in base class BaseMixture.

But it seems to be necessary because part of the initialization is in BaseMixture . This variable will be set to be True or False in subclass GaussianMixture, and so I need one variable in base to know if the base class's fit_transform can skip the initialization.