Fixed bug in esvd for m>n

fluents/lib/engines.py

@@ -8,6 +8,11 @@ from scipy.linalg import svd,inv
 from scipy import dot,empty,eye,newaxis,zeros,sqrt,diag,\
      apply_along_axis,mean,ones,randn,empty_like,outer,c_,\
      rand,sum,cumsum,matrix
+has_sym=True
+try:
+    import symmeig
+except:
+    has_sym = False
 
 def pca(a, aopt, scale='scores', mode='normal'):
     """ Principal Component Analysis model
@@ -18,11 +23,11 @@ def pca(a, aopt, scale='scores', mode='normal'):
     """
     
     m, n = a.shape
-
-    if m*3>n:
+    #print "rows: %s    cols: %s" %(m,n)
+    if m>(n+100) or n>(m+100):
         u, s, v = esvd(a)
     else:
-        u, s, vt = svd(a, full_matrices=0)
+        u, s, vt = svd(a, 0)
         v = vt.T
     eigvals = (1./m)*s
     T = u*s
@@ -250,6 +255,7 @@ def nipals_lpls(X, Y, Z, a_max, alpha=.7, mean_ctr=[2, 0, 1], mode='normal', sca
         X, mnX = center(X, xctr)
         Y, mnY = center(Y, xctr)
         Z, mnZ = center(Z, zctr)
+        print Z.mean(1)
 
     varX = pow(X, 2).sum()
     varY = pow(Y, 2).sum()
@@ -343,28 +349,29 @@ def nipals_lpls(X, Y, Z, a_max, alpha=.7, mean_ctr=[2, 0, 1], mode='normal', sca
 def m_shape(array):
     return matrix(array).shape
 
-def esvd(data, economy=1):
+def esvd(data):
     """SVD with the option of economy sized calculation
     Calculate subspaces of X'X or XX' depending on the shape
     of the matrix.
 
     Good for extreme fat or thin matrices
 
+    :notes:
     Numpy supports this by setting full_matrices=0
     """
     m, n = data.shape
     if m>=n:
-        data = dot(data.T, data)
-        u, s, vt = svd(data)
+        kernel = dot(data.T, data)
+        u, s, vt = svd(kernel)
         u = dot(data, vt.T)
         v = vt.T
         for i in xrange(n):
             s[i] = vnorm(u[:,i])
             u[:,i] = u[:,i]/s[i]
     else:
-        data = dot(data, data.T)
-        data = (data + data.T)/2.0
-        u, s, vt = svd(data)
+        kernel = dot(data, data.T)
+        #data = (data + data.T)/2.0
+        u, s, vt = svd(kernel)
         v = dot(u.T, data)
         for i in xrange(m):
             s[i] = vnorm(v[i,:])