Estimating the parameters of gaussians to fit a lot of samples can be do with Exceptation Maximization, for instance if we want to fit two gaussian on points, to have the clusters a and b. (1) b i = P ( b | x i ) = P ( x i | b ) P ( b ) P ( x i | b ) P ( b ) + P ( x i | b ) P ( a ) Here P ( b ) is the prior that depicts the overall importance of the b cluster.This prior is then updated for the next step, according on how many the points belongs to the b cluster: P ( b ) = 1 n ∑ i b i However, what is the value of the prior P ( b ) on the first iteration of the algorithm?

Question

Estimating the parameters of gaussians to fit a lot of samples can be do with Exceptation Maximization, for instance if we want to fit two gaussian on points, to have the clusters   a and   b. (1)      b    i    =  P  (  b      |        x    i    )  =            P      (              x        i                    |            b      )      P      (      b      )              P      (              x        i                    |            b      )      P      (      b      )      +      P      (              x        i                    |            b      )      P      (      a      )      Here   P  (  b  ) is the prior that depicts the overall importance of the   b cluster.This prior is then updated for the next step, according on how many the points belongs to the   b cluster:  P  (  b  )  =      1    n        ∑    i        b    i  However, what is the value of the prior   P  (  b  ) on the first iteration of the algorithm?

Sophia Mcdowell · Accepted Answer

Some simple suggestions: you could initialize with k-means and/or run multiple tries with different random initialization. Note that it is possible for EM algorithms to get stuck in local minima, so multiple initializations can be useful.

Estimating the parameters of gaussians to fit a lot of samples can be do with Exceptation Maximizati

Answered question

Answer & Explanation

New Questions in High school geometry