K-Means - Understanding Loss, Coordinate Ascent

#1

I am not getting how one time we are fixing uk and updating r and one time we are fixing r and updating uk
plz explain.

#2

K-means clustering is solved using coordinate descent where we have 2 independent parameters variables Uk and Rnk.
The idea behind Coordinate descent is that the minimization of a multivariable function (loss/inertia) can be achieved by minimizing it along one direction at a time, i.e., solving univariate, so we kept one parameter variable and other constant alternatively.

E-step : We find Rnk keeping Uk as fixed
M-step : We find Uk keeping Rnk as fixed

Hope this helped :slight_smile:

#3

ya This really helped .Thanks…