Update

Author: phillity

EM/EM.py

@@ -123,7 +123,7 @@ def SSE(C, mu):
 if len(D.shape) < 2:
  D = D.reshape((D.shape[0],1))
 
-# Read in k centroid count
+# Read in k cluster count
 k = int(sys.argv[2])
 
 # Read in mu mean matrix if given

SpectralClustering/kMeans.py

@@ -55,16 +55,9 @@ def kMeans(D, k, mu=None, eps=0.0001):
  mu[i,:] = np.sum(C[i], axis=0) / len(C[i])
 
  # Check for convergence
- check = np.linalg.norm(mu - prev_mu)
- if check <= eps:
+ if np.linalg.norm(mu - prev_mu) <= eps:
  return C, labels, mu, iter
 
- # Print update
- #print("Iteration " + str(iter) + ":")
- #for i in range(k):
- #print("c_" + str(i) + ":" + str(C[i]) + " mu_" + str(i) + ":" + str(mu[i,:]))
- #print("||mu - mu_prev||:" + str(check) + "\n") 
-
  # Update iteration count
  iter += 1
  # Update previous mu

datafile_1d.txt

@@ -0,0 +1,9 @@
+2
+3
+4
+10
+11
+12
+20
+25
+30

kMeans/kMeans.py

@@ -11,13 +11,14 @@
 # mu - kxd centroid matrix
 # k - cluster count
 # eps - convergence tolarence
-def kMeans(D, k, mu, eps=0.0001):
+def kMeans(D, k, mu=None, eps=0.0001):
  # Get dimensions of nxd D matrix
  n, d = D.shape 
 
  # If mu is not preset
- if np.array_equal(mu,np.zeros((k,d))):
+ if mu is None:
  # Randomly intialize k centroids in kxd mu matrix
+ mu = np.zeros((k,d))
  mu_list = np.random.choice(n,size=k,replace=False)
  for i,id in enumerate(mu_list):
  mu[i,:] = D[id,:]
@@ -55,13 +56,7 @@ def kMeans(D, k, mu, eps=0.0001):
 
  # Check for convergence
  if np.linalg.norm(mu - prev_mu) <= eps:
- return C, labels, mu, iter
-
- # Print update
- #print("Iteration " + str(iter) + ":")
- #for i in range(k):
- #print("c_" + str(i) + ":" + str(C[i]) + " mu_" + str(i) + ":" + str(mu[i,:]))
- #print("||mu - mu_prev||:" + str(check) + "\n") 
+ return C, labels, mu, iter 
 
  # Update iteration count
  iter += 1
@@ -98,7 +93,7 @@ def SSE(C, mu):
 if len(D.shape) < 2:
  D = D.reshape((D.shape[0],1))
 
-# Read in k centroid count
+# Read in k cluster count
 k = int(sys.argv[2])
 
 # Read in mu centroid id list if given