main.py

import numpy as np
import kmeans
import common
import naive_em
import em
from matplotlib import pyplot as plt

X = np.loadtxt("toy_data.txt")

K = [1,2,3,4]
k = len(K)
seeds = [0,1,2,3,4]
mixtures_k, posts_k, costs_k, seeds_k, mixtures_em, posts_em, costs_em, seeds_em = k*[[]], k*[[]], k*[10000], k*[0], k*[[]], k*[[]], k*[-10000000], k*[0]

#===============================
#       (2) K-Means
#===============================

print('=================/n   K-MEANS\n=================  ')

for k in K:
  for seed in seeds:
    mix, post = common.init(X, k, seed) # random initialization
    mix_k, post_k, cost_k = kmeans.run(X, mix, post) # run kmeans
    # Only keep max values
    print(f'k={k}, seed={seed}, cost_k={cost_k}')
    if cost_k < costs_k[k-1]:
       mixtures_k[k-1] = mix_k
       posts_k[k-1] = post_k
       costs_k[k-1] = cost_k
       seeds_k[k-1] = seed
#  common.plot(X, mixtures_k[k-1], posts_k[k-1],f'K-means K={k}, seed={seeds_k[k-1]}')

#===============================
#   (4) K-means vs. EM
#===============================

print('=================/n   EM\n=================  ')
mixtures_k, posts_k, costs_k, seeds_k = k*[[]], k*[[]], k*[10000], k*[0]

for k in K:
  for seed in seeds:
    mix, post = common.init(X, k, seed) # random initialization
    mix_em, post_em, cost_em = naive_em.run(X, mix, post)
    # Only keep max values
    print(f'k={k}, seed={seed}, cost_k={cost_k}')
    if cost_em > costs_em[k-1]:
        mixtures_em[k-1] = mix_em
        posts_em[k-1] = post_em
        costs_em[k-1] = cost_em
        seeds_em[k-1] = seed
#  common.plot(X, mixtures_k[k-1], posts_k[k-1],f'Naive EM K={k}, seed={seeds_k[k-1]}')


# ============================
#       (5) BIC
# ============================

print('=================/n   BAYESIAN CRITERION\n=================  ')

k = len(K)
bic = k*[0]
for k in K:
  bic[k-1] = common.bic(X, mixtures_em[k-1], costs_em[k-1])

plt.plot(K, bic)
plt.show()
print(bic)