Mixture of Linear Experts does not appear to use EM

[dfd]

https://github.com/probml/pyprobml/blob/master/notebooks/book1/13/mixexpDemoOneToMany.ipynb

The random initialization of parameters and weights occurs within the E-Step (instead of before it), so it's not really making use of the fit parameters from the M-Step in each subsequent iteration.

      #E-step :
      np.random.seed(iteration)
      Wy = 0.1*np.random.randn(D, K)
      bias = 0.3*np.random.randn(D, K)
      mixweights = np.random.rand(1, K)
      normmw =  np.linalg.norm(mixweights)
      mixweights = mixweights/normmw
      sigma2 = 0.1*np.random.randn(1, K)

[berkott]

I observed this same issue today. Running with max_iter = n and iteration = 0 gives the same result as max_iter = n and iteration = n-1.