Port to python3

pymf/aa.py

@@ -11,8 +11,8 @@
 import numpy as np
 from cvxopt import solvers, base
 
-from svd import pinv
-from base import PyMFBase
+from pymf.svd import pinv
+from pymf.base import PyMFBase
 __all__ = ["AA"]
 
 class AA(PyMFBase):
@@ -104,7 +104,7 @@ def update_single_h(i):
         INQb = base.matrix(0.0, (self._num_bases,1))
         EQa = base.matrix(1.0, (1, self._num_bases))
 
-        for i in xrange(self._num_samples):
+        for i in range(self._num_samples):
             update_single_h(i)        
 
     def _update_w(self):
@@ -126,7 +126,7 @@ def update_single_w(i):
         INQb = base.matrix(0.0, (self._num_samples, 1))
         EQa = base.matrix(1.0, (1, self._num_samples))
 
-        for i in xrange(self._num_bases):
+        for i in range(self._num_bases):
             update_single_w(i)            
 
         self.W = np.dot(self.beta, self.data.T).T

pymf/base.py

@@ -248,13 +248,13 @@ def factorize(self, niter=100, show_progress=False,
         if compute_err:
             self.ferr = np.zeros(niter)
 
-        for i in xrange(niter):
+        for i in range(niter):
             if compute_w:
                 self._update_w()
 
             if compute_h:
-                self._update_h()                                        
-         
+                self._update_h()
+
             if compute_err:                 
                 self.ferr[i] = self.frobenius_norm()                
                 self._logger.info('FN: %s (%s/%s)'  %(self.ferr[i], i+1, niter))

pymf/bnmf.py

@@ -9,10 +9,11 @@
 Applications. ICDM 2007
 """
 import numpy as np
-from base import PyMFBase
+from pymf.base import PyMFBase
 
 __all__ = ["BNMF"]
 
+
 class BNMF(PyMFBase):
     """      
     BNMF(data, data, num_bases=4)

pymf/chnmf.py

@@ -12,9 +12,9 @@
 import numpy as np
 
 from itertools import combinations
-from dist import vq
-from pca import PCA
-from aa import AA
+from pymf.dist import vq
+from pymf.pca import PCA
+from pymf.aa import AA
 
 __all__ = ["CHNMF"]
 
@@ -171,7 +171,7 @@ def select_hull_points(data, n=3):
             pcamodel = PCA(self.data)        
             pcamodel.factorize(show_progress=False)        
             proj = pcamodel.H
-            print "PROJ", proj.shape
+            print("PROJ", proj.shape)
         else:            
             R = np.random.randn(self._base_sel, self._data_dimension)           
             proj = np.dot(R, self.data)

pymf/cmde.py

@@ -10,7 +10,7 @@
 Mining. 
 """
 import numpy as np
-from cur import CUR
+from pymf.cur import CUR
 
 __all__ = ["CMD"]
 

pymf/cmeans.py

@@ -8,8 +8,8 @@
 """
 import numpy as np
 
-import dist
-from base import PyMFBase
+from pymf.dist import pdist
+from pymf.base import PyMFBase
 
 __all__ = ["Cmeans"]
 
@@ -65,7 +65,7 @@ class Cmeans(PyMFBase):
     def _update_h(self):                    
         # assign samples to best matching centres ...
         m = 1.75
-        tmp_dist = dist.pdist(self.W, self.data, metric='l2') + self._EPS
+        tmp_dist = pdist(self.W, self.data, metric='l2') + self._EPS
         self.H[:,:] = 0.0
 
         for i in range(self._num_bases):

pymf/cnmf.py

@@ -10,8 +10,8 @@
 """
 import numpy as np
 import logging
-from kmeans import Kmeans
-from base import PyMFBase
+from pymf.kmeans import Kmeans
+from pymf.base import PyMFBase
 
 __all__ = ["CNMF"]
 
@@ -143,7 +143,7 @@ def separate_negative(m):
         self.ferr = np.zeros(niter)
         # iterate over W and H
 
-        for i in xrange(niter):
+        for i in range(niter):
             # update H
             XtX_neg_x_W = np.dot(XtX_neg, self.G)
             XtX_pos_x_W = np.dot(XtX_pos, self.G)

pymf/cur.py

@@ -11,8 +11,8 @@
 import numpy as np
 import scipy.sparse
 
-from svd import pinv
-from base import PyMFBase3
+from pymf.svd import pinv
+from pymf.base import PyMFBase3
 
 
 __all__ = ["CUR"]

pymf/cursl.py

@@ -11,8 +11,8 @@
 import numpy as np
 import scipy.sparse
 
-from svd import pinv, SVD
-from cmde import CMD
+from pymf.svd import SVD
+from pymf.cmde import CMD
 
 __all__ = ["CURSL"]
 

pymf/dist.py

@@ -106,13 +106,13 @@ def pdist(A, B, metric='l2' ):
     # Returns a distance matrix d (n x m).
     d = np.zeros((A.shape[1], B.shape[1]))
     if A.shape[1] <= B.shape[1]:
-        for aidx in xrange(A.shape[1]):
+        for aidx in range(A.shape[1]):
             if metric == 'l2':
                 d[aidx:aidx+1,:] = l2_distance(B[:,:], A[:,aidx:aidx+1]).reshape((1,-1))
             if metric == 'l1':
                 d[aidx:aidx+1,:] = l1_distance(B[:,:], A[:,aidx:aidx+1]).reshape((1,-1))
     else:
-        for bidx in xrange(B.shape[1]):
+        for bidx in range(B.shape[1]):
             if metric == 'l2':
                 d[:, bidx:bidx+1] = l2_distance(A[:,:], B[:,bidx:bidx+1]).reshape((-1,1))
             if metric == 'l1':

pymf/greedy.py

@@ -11,8 +11,8 @@
 import time
 import scipy.sparse
 import numpy as np
-from svd import *
-from base import PyMFBase
+from pymf.svd import pinv, SVD
+from pymf.base import PyMFBase
 
 __all__ = ["GREEDY"]
 

pymf/greedycur.py

@@ -11,8 +11,8 @@
 Reconstruction via Greedy Approximation of SVD. ISAAC'2008.
 """
 import numpy as np
-from greedy import GREEDY
-from cur import CUR
+from pymf.greedy import GREEDY
+from pymf.cur import CUR
 
 __all__ = ["GREEDYCUR"]
 
@@ -66,7 +66,7 @@ def sample(self, A, c):
         """
         # set k to a value lower than the number of bases, usually
         # gives better results.
-        k = np.round(c - c/5.0)
+        k = np.rint(c - c/5.0).astype(np.int64)
         greedy_mdl = GREEDY(A, k=k, num_bases=c)
         greedy_mdl.factorize(compute_h=False, compute_err=False, niter=1)        
         return greedy_mdl.select

pymf/kmeans.py

@@ -6,8 +6,8 @@
 import numpy as np
 import random
 
-import dist
-from base import PyMFBase
+from pymf.base import PyMFBase
+from pymf.dist import vq
 
 __all__ = ["Kmeans"]
 
@@ -61,15 +61,15 @@ def _init_h(self):
 
     def _init_w(self):
         # set W to some random data samples
-        sel = random.sample(xrange(self._num_samples), self._num_bases)
+        sel = random.sample(range(self._num_samples), self._num_bases)
 
         # sort indices, otherwise h5py won't work
         self.W = self.data[:, np.sort(sel)]        
 
 
     def _update_h(self):                    
         # and assign samples to the best matching centers
-        self.assigned = dist.vq(self.W, self.data)
+        self.assigned = vq(self.W, self.data)
         self.H = np.zeros(self.H.shape)
         self.H[self.assigned, range(self._num_samples)] = 1.0
 

pymf/laesa.py

@@ -9,11 +9,9 @@
 neighbour approximating and eliminating search algorithm with linear 
 preprocessing-time and memory requirements. Pattern Recognition Letters 1994.
 """
-import scipy.sparse
 import numpy as np
 
-from dist import *
-from sivm import SIVM
+from pymf.sivm import SIVM
 
 __all__ = ["LAESA"]
 

pymf/nmf.py

@@ -14,8 +14,7 @@
 import scipy.sparse
 import scipy.optimize
 from cvxopt import solvers, base
-from base import PyMFBase
-from svd import pinv
+from pymf.base import PyMFBase
 
 __all__ = ["NMF", "RNMF", "NMFALS", "NMFNNLS"]
 
@@ -66,14 +65,14 @@ class NMF(PyMFBase):
 
     def _update_h(self):
         # pre init H1, and H2 (necessary for storing matrices on disk)
-        H2 = np.dot(np.dot(self.W.T, self.W), self.H) + 10**-9
-        self.H *= np.dot(self.W.T, self.data[:,:])
+        H2 = ((self.W.T @ self.W) @ self.H) + 10**-9
+        self.H *= self.W.T @ self.data
         self.H /= H2
 
     def _update_w(self):
         # pre init W1, and W2 (necessary for storing matrices on disk)
-        W2 = np.dot(np.dot(self.W, self.H), self.H.T) + 10**-9
-        self.W *= np.dot(self.data[:,:], self.H.T)
+        W2 = ((self.W @ self.H) @ self.H.T) + 10**-9
+        self.W *= (self.data @ self.H.T)
         self.W /= W2
         self.W /= np.sqrt(np.sum(self.W**2.0, axis=0))
 
@@ -134,8 +133,7 @@ def soft_thresholding(self, X, lamb):
         return X
 
     def _init_h(self):
-        self.H = np.random.random((self._num_bases, self._num_samples))
-        self.H[:,:] = 1.0
+        self.H = np.ones((self._num_bases, self._num_samples), dtype=np.float64)
 
         # normalized bases
         Wnorm = np.sqrt(np.sum(self.W**2.0, axis=0))
@@ -146,7 +144,7 @@ def _init_h(self):
 
         self._update_s()
 
-    def _update_s(self):                
+    def _update_s(self):
         self.S = self.data - np.dot(self.W, self.H)
         self.S = self.soft_thresholding(self.S, self._lamb)
 
@@ -228,7 +226,7 @@ def updatesingleH(i):
         INQa = base.matrix(-np.eye(self._num_bases))
         INQb = base.matrix(0.0, (self._num_bases,1))            
 
-        map(updatesingleH, xrange(self._num_samples))                        
+        list(map(updatesingleH, range(self._num_samples)))
 
 
     def _update_w(self):
@@ -243,7 +241,7 @@ def updatesingleW(i):
         INQa = base.matrix(-np.eye(self._num_bases))
         INQb = base.matrix(0.0, (self._num_bases,1))            
 
-        map(updatesingleW, xrange(self._data_dimension))
+        list(map(updatesingleW, range(self._data_dimension)))
 
         self.W = self.W/np.sum(self.W, axis=1)
 
@@ -298,14 +296,14 @@ def _update_h(self):
         def updatesingleH(i):        
             self.H[:,i] = scipy.optimize.nnls(self.W, self.data[:,i])[0]
 
-        map(updatesingleH, xrange(self._num_samples))                        
+        list(map(updatesingleH, range(self._num_samples)))
 
 
     def _update_w(self):
         def updatesingleW(i):            
             self.W[i,:] = scipy.optimize.nnls(self.H.T, self.data[i,:].T)[0]
 
-        map(updatesingleW, xrange(self._data_dimension))
+        list(map(updatesingleW, range(self._data_dimension)))
 
 
 def _test():

pymf/nndsvd.py

@@ -10,9 +10,9 @@
 """
 import numpy as np
 
-from base import PyMFBase
-from svd import SVD
-from nmf import NMF
+from pymf.base import PyMFBase
+from pymf.svd import SVD
+from pymf.nmf import NMF
 
 __all__ = ["NNDSVD"]
 

pymf/pca.py

@@ -7,8 +7,8 @@
 """
 import numpy as np
 
-from base import PyMFBase
-from svd import SVD
+from pymf.base import PyMFBase
+from pymf.svd import SVD
 
 
 __all__ = ["PCA"]

pymf/sivm.py

@@ -12,11 +12,12 @@
 import scipy.sparse
 import numpy as np
 
-from dist import *
-from aa import AA
+from pymf.dist import l1_distance, l2_distance, cosine_distance, abs_cosine_distance, kl_divergence, weighted_abs_cosine_distance
+from pymf.aa import AA
 
 __all__ = ["SIVM"]
 
+
 class SIVM(AA):
     """      
     SIVM(data, num_bases=4, dist_measure='l2')

pymf/sivm_cur.py

@@ -10,12 +10,12 @@
 Conf. on Information and Knowledge Management. ACM. 2010.
 """
 import numpy as np
-import scipy
-from sivm import SIVM
-from cur import CUR
+from pymf.sivm import SIVM
+from pymf.cur import CUR
 
 __all__ = ["SIVM_CUR"]
 
+
 class SIVM_CUR(CUR):
     '''
     SIVM_CUR(data, num_bases=4, dist_measure='l2')

pymf/sivm_gsat.py

@@ -160,7 +160,7 @@ def factorize(self, show_progress=False, compute_w=True, compute_h=True,
         if compute_err:
             self.ferr = np.zeros(niter)
 
-        for i in xrange(niter):
+        for i in range(niter):
             if compute_w:
                 self._update_w()
 

pymf/sivm_search.py

@@ -12,9 +12,9 @@
 import scipy.sparse
 import numpy as np
 
-from dist import pdist
+from pymf.dist import pdist
 from vol import *
-from sivm import SIVM
+from pymf.sivm import SIVM
 
 __all__ = ["SIVM_SEARCH"]