LBG algo

455-Codage_Sources/algo_code/LBG.py

@@ -0,0 +1,61 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Sun May  5 13:59:37 2019
+
+@author: pac
+Algorithme de Linde-Buzo-Gray, version 2D
+"""
+
+import numpy as np
+import matplotlib.pyplot as plt
+
+from scipy.spatial import Voronoi, voronoi_plot_2d
+
+mean= [0,0]
+cov = [[1,0],[0,1]]
+M = 20;
+N =100; #point par cluster
+K = N*M
+means = np.random.rand(M,2)*10
+X = np.zeros((K,2))
+plt.figure()
+
+for m in range(M):
+    xi = np.random.multivariate_normal(means[m,:],cov,N)
+    X[m*N:(m+1)*N] = xi
+    plt.plot(xi[:,0],xi[:,1],'+')
+plt.plot(means[:,0],means[:,1],'ob')
+plt.show()
+
+# X = np.random.multivariate_normal(mean,cov,K)
+Y0 = np.random.multivariate_normal(mean, cov,M)
+Y0 = means;
+def LBG(X,Y0,eps=1e-5,maxiter=1000):
+    Y = Y0.copy()
+    old_dist = np.inf
+    cluster_index = np.zeros(K,dtype=int)
+    for l in range(maxiter):
+        dist= 0;
+        for k in range(len(X)):
+            quant_min =np.inf
+            for j in range(len(Y)):
+                if np.linalg.norm(X[k]-Y[j]) <np.linalg.norm(X[k]-quant_min):
+                    quant_min = Y[j]
+                    cluster_index[k] = j
+            dist += sum((X[k]-quant_min)**2)
+        for j in range(len(Y)):
+            Y[j,:] = np.mean(X[cluster_index==j],axis=0)            
+        if dist-old_dist < eps:
+            break
+        else:
+            old_dist = dist
+    return Y
+Y = LBG(X,Y0)
+vor = Voronoi(Y) 
+voronoi_plot_2d(vor,show_vertices=False)
+print(Y)
+plt.plot(X[:,0],X[:,1],'+')
+plt.plot(Y[:,0],Y[:,1],'o')           
+#plt.plot(Y0[:,0],Y0[:,1],'ob')  
+plt.show()