We do not update a cluster if it is not associated to any data point.

quillotm
1 parent 4152e83df2
Showing 1 changed file with 3 additions and 3 deletions Side-by-side Diff
volia/clustering_modules/kmeans_mahalanobis.py
@@ -27,7 +27,6 @@
         for n in range(N):
             for k in range(self.K):
                 distances[n][k] = self._dist(features[n], self.C[k], self.L[k])
-        print(distances)
         closest_cluster = np.argmin(distances, axis=1)
         return closest_cluster
  
@@ -119,7 +118,6 @@
             for n in range(N):
                 for k in range(self.K):
                     distances[n][k] = self._dist(X[n], self.C[k], self.L[k])
-            print(distances)
             closest_cluster = np.argmin(distances, axis=1)
             if i % 1 == 0:
                 # -- Debug tool ----------------------
@@ -140,6 +138,8 @@
                 # Find subset of X with values closed to the centroid c_k.
                 X_sub = np.where(closest_cluster == k)
                 X_sub = np.take(X, X_sub[0], axis=0)
+                if X_sub.shape[0] == 0:
+                    continue
                 np.mean(X_sub, axis=0)
                 C_new = np.mean(X_sub, axis=0)
  
@@ -150,7 +150,7 @@
                     c_tmp = np.reshape(C_new, (-1, 1))
                     K_new = K_new + (x - c_tmp).dot((x - c_tmp).transpose())
                 K_new = K_new / X_sub.shape[0]
-                K_new = np.linalg.inv(K_new)
+                K_new = np.linalg.pinv(K_new)
  
                 if end_algo and (not (self.C[k] == C_new).all()):  # If the same stop
                     end_algo = False
...	...	@@ -27,7 +27,6 @@
27	27	for n in range(N):
28	28	for k in range(self.K):
29	29	distances[n][k] = self._dist(features[n], self.C[k], self.L[k])
30		- print(distances)
31	30	closest_cluster = np.argmin(distances, axis=1)
32	31	return closest_cluster
33	32
...	...	@@ -119,7 +118,6 @@
119	118	for n in range(N):
120	119	for k in range(self.K):
121	120	distances[n][k] = self._dist(X[n], self.C[k], self.L[k])
122		- print(distances)
123	121	closest_cluster = np.argmin(distances, axis=1)
124	122	if i % 1 == 0:
125	123	# -- Debug tool ----------------------
...	...	@@ -140,6 +138,8 @@
140	138	# Find subset of X with values closed to the centroid c_k.
141	139	X_sub = np.where(closest_cluster == k)
142	140	X_sub = np.take(X, X_sub[0], axis=0)
	141	+ if X_sub.shape[0] == 0:
	142	+ continue
143	143	np.mean(X_sub, axis=0)
144	144	C_new = np.mean(X_sub, axis=0)
145	145
...	...	@@ -150,7 +150,7 @@
150	150	c_tmp = np.reshape(C_new, (-1, 1))
151	151	K_new = K_new + (x - c_tmp).dot((x - c_tmp).transpose())
152	152	K_new = K_new / X_sub.shape[0]
153		- K_new = np.linalg.inv(K_new)
	153	+ K_new = np.linalg.pinv(K_new)
154	154
155	155	if end_algo and (not (self.C[k] == C_new).all()): # If the same stop
156	156	end_algo = False