I've reading recently on Expectation Maximization (EM) and it keeps coming up that Initializing EM using K-Means is a good idea but i'm having difficulties in grasping this notion.
So as far as i know when using kmeans, the result you get is coordinates of the clusters' centroids according to the pre-defined numberof clusters, so how can this be used in order to initialize EM. To make things clearer this is the problem i'm currently trying to solve:
I have a dataset of noisy data points Y who originates from Samples X taken from an 8-ASK set. Now i loaded my dataset and have used a kmeans algorithm in order to identify the centroids but can't seem to know what's the next step. The EM algorithm that i use requires the parameters: the initial start values for the centroids and their probability distribution as well as the initial mean and variance but i do not understand how can get those exactly.
To summarize my question is basically how can i calculate the mean, variance and initial diribution of the centroids generated by kmeans algorithm when i ran him on my data Y ?
