Cleaned esvd routine, added subfunc scale

fluents/lib/engines.py

@@ -7,7 +7,7 @@ import math
 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,r_,c_,\
-     rand,sum,cumsum,matrix, expand_dims,minimum,where
+     rand,sum,cumsum,matrix, expand_dims,minimum,where,arange
 has_sym=True
 try:
     from symeig import symeig
@@ -67,8 +67,8 @@ def pca(a, aopt,scale='scores',mode='normal',center_axis=0):
     if center_axis>=0:
         a = a - expand_dims(a.mean(center_axis), center_axis)
     if m>(n+100) or n>(m+100):
-        u, e, v = esvd(a, amax=None) # fixme:amax option need to work with expl.var
-        s = sqrt(e)
+        u, s, v = esvd(a, amax=None) # fixme:amax option need to work with expl.var
+        print s[:10]
     else:
         u, s, vt = svd(a, 0)
         v = vt.T
@@ -189,7 +189,7 @@ def pcr(a, b, aopt, scale='scores',mode='normal',center_axis=0):
     dat.update({'Q':Q, 'F':F, 'expvary':expvary})
     return dat
 
-def pls(a, b, aopt=2, scale='scores', mode='normal', center_axis=0, ab=None):
+def pls(a, b, aopt=2, scale='scores', mode='normal', center_axis=-1, ab=None):
     """Partial Least Squares Regression.
 
     Performs PLS on given matrix and returns results in a dictionary.
@@ -696,34 +696,38 @@ def esvd(data, amax=None):
     if m>=n:
         kernel = dot(data.T, data)
         if has_sym:
-            if not amax:
-                amax = n-1
+            if amax==None:
+                amax = n
+                pcrange = None
+            else:
                 pcrange = [n-amax, n]
+            print "symm>n"
             s, v = symeig(kernel, range=pcrange, overwrite=True)
             s = s[::-1]
-            v = v[:,arange(n, -1, -1)]
+            v = v[:,::-1]
         else:
             u, s, vt = svd(kernel)
             v = vt.T
-        u = dot(data, v)
-        for i in xrange(amax):
-            s[i] = vnorm(u[:,i])
-            u[:,i] = u[:,i]/s[i]
+        s = sqrt(s)
+        u = dot(data, v)/s
     else:
         kernel = dot(data, data.T)
         if has_sym:
-            if not amax:
-                amax = m-1
+            if amax==None:
+                amax = m
+                pcrange = None
+            else:
                 pcrange = [m-amax, m]
+            print "sym (m<n)"
             s, u = symeig(kernel, range=pcrange, overwrite=True)
+            s = s[::-1]
+            u = u[:,::-1]
         else:
             u, s, vt = svd(kernel)
-        v = dot(u.T, data)
-        for i in xrange(amax):
-            s[i] = vnorm(v[i,:])
-            v[i,:] = v[i,:]/s[i]
-
-    return u, s, v.T
+        s = sqrt(s)
+        v = dot(data.T, u)/s
+    print s[:2]
+    return u, s, v
 
 def vnorm(x):
     # assume column arrays (or vectors)
@@ -744,3 +748,16 @@ def center(a, axis):
         raise IOError("input error: axis must be in [-1,0,1,2]")
 
     return a - mn, mn
+
+def scale(a, axis):
+    if axis==-1:
+        sc = zeros((a.shape[1],))
+    elif axis==0:
+        sc = a.std(0)
+    elif axis==1:
+        sc = a.std(1)[:,newaxis]
+    else:
+        raise IOError("input error: axis must be in [-1,0,1]")
+
+    return a - sc, sc
+