Multiple lib changes

fluents/lib/engines.py

@@ -1,4 +1,3 @@
-
 """Module contain algorithms for  (burdensome) calculations.
 
 There is no typechecking of any kind here, just focus on speed
@@ -7,7 +6,7 @@ There is no typechecking of any kind here, just focus on speed
 from scipy.linalg import svd,norm,inv,pinv,qr
 from scipy import dot,empty,eye,newaxis,zeros,sqrt,diag,\
      apply_along_axis,mean,ones,randn,empty_like,outer,c_,\
-     rand,sum,cumsum
+     rand,sum,cumsum,matrix
 
 def pca(a, aopt, scale='scores', mode='normal'):
     """ Principal Component Analysis model
@@ -17,8 +16,9 @@ def pca(a, aopt, scale='scores', mode='normal'):
          -- detailed    : returns all model params and all residuals
     """
     
-    m,n = a.shape
-    u,s,vt = svd(a, full_matrices=0)
+    m, n = a.shape
+    u, s, vt = svd(a, full_matrices=0)
+    eigvals = (1./m)*s
     T = u*s
     T = T[:,:aopt]
     P = vt[:aopt,:].T
@@ -43,17 +43,40 @@ def pca(a, aopt, scale='scores', mode='normal'):
             
     return {'T':T, 'P':P, 'E':E}
 
+def pcr(a, b, aopt=2, scale='scores', mode='normal'):
+    """Returns Principal component regression model."""
+    m, n = a.shape
+    try:
+        k, l = b.shape
+    except:
+        k = b.shape[0]
+        l = 1
+    B = empty((aopt, n, l))
+    U, s, Vt = svd(a, full_matrices=True)
+    T = U*s
+    T = T[:,:aopt]
+    P = Vt[:aopt,:].T
+    Q = dot(dot(inv(dot(T.T, T)), T.T), b).T
+    for i in range(aopt):
+        ti = T[:,:i+1]
+        r = dot(dot(inv(dot(ti.T,ti)), ti.T), b)
+        B[i] = dot(P[:,:i+1], r)
+    E = a - dot(T, P.T)
+    F = b - dot(T, Q.T)
+
+    return {'T':T, 'P':P,'Q': Q, 'B':B, 'E':E, 'F':F}
+    
 def pls(a, b, aopt=2, scale='scores', mode='normal', ab=None):
     """Kernel pls for tall/wide matrices.
 
     Fast pls for calibration. Only inefficient for many Y-vars.
     
     """
-    m,n = a.shape
+    m, n = a.shape
     if ab!=None:
-        mm,l = ab.shape
+        mm, l = m_shape(ab)
     else:
-        k,l = b.shape
+        k, l = m_shape(b)
 
     W = empty((n, aopt))
     P = empty((n, aopt))
@@ -66,10 +89,10 @@ def pls(a, b, aopt=2, scale='scores', mode='normal', ab=None):
         ab = dot(a.T, b)
     for i in range(aopt):
         if ab.shape[1]==1:
-            w = ab
+            w = ab.reshape(mm, l)
         else:
-            u,s,vh = svd(dot(ab.T, ab))
-            w = dot(ab,u[:,:1])
+            u, s, vh = svd(dot(ab.T, ab))
+            w = dot(ab, u[:,:1])
     
         w = w/norm(w)
         r = w.copy()
@@ -99,9 +122,9 @@ def pls(a, b, aopt=2, scale='scores', mode='normal', ab=None):
     if mode=='detailed':
         E = empty((aopt, m, n))
         F = empty((aopt, k, l))
-        for i in range(1,aopt+1,1):
-            E[i-1] = a - dot(T[:,:i],P[:,:i].T)
-            F[i-1] = b - dot(T[:,:i],Q[:,:i].T)
+        for i in range(1, aopt+1, 1):
+            E[i-1] = a - dot(T[:,:i], P[:,:i].T)
+            F[i-1] = b - dot(T[:,:i], Q[:,:i].T)
     else:
         E = a - dot(T[:,:aopt], P[:,:aopt].T)
         F = b - dot(T[:,:aopt], Q[:,:aopt].T)
@@ -121,17 +144,17 @@ def w_simpls(aat, b, aopt):
     There is no P,W.  T is normalised
     """
     bb = b.copy()
-    m,m = aat.shape
+    m, m = aat.shape
     U = empty((m, aopt))
     T = empty((m, aopt))
     H = empty((m, aopt)) #just like W in simpls
     PROJ = empty((m, aopt)) #just like R in simpls
 
     for i in range(aopt):
-        u,s,vh = svd(dot(dot(b.T, aat), b), full_matrices=0)
+        u, s, vh = svd(dot(dot(b.T, aat), b), full_matrices=0)
         u = dot(b, u[:,:1]) #y-factor scores
         U[:,i] = u.ravel()
-        t =dot(aat, u)
+        t = dot(aat, u)
         t = t/norm(t)
         T[:,i] = t.ravel()
         h = dot(aat, t) #score-weights
@@ -141,19 +164,18 @@ def w_simpls(aat, b, aopt):
             b = b - dot(PROJ[:,:i+1], dot(H[:,:i+1].T,b) )
     C = dot(bb.T, T)
 
-    return {'T':T,'U':U,'Q':C,'H':H}
+    return {'T':T, 'U':U, 'Q':C, 'H':H}
 
 def bridge(a, b, aopt, scale='scores', mode='normal', r=0):
     """Undeflated Ridged svd(X'Y)
     """
     m, n = a.shape
-    k, l = b.shape
-    u,s,vt = svd(b, full_matrices=0)
+    k, l = m_shape(b)
+    u, s, vt = svd(b, full_matrices=0)
     g0 = dot(u*s, u.T)
     g = (1 - r)*g0 + r*eye(m)
     ag = dot(a.T, g)
-    
-    u,s,vt = svd(ag, full_matrices=0)
+    u, s, vt = svd(ag, full_matrices=0)
     W = u[:,:aopt]
     K = vt[:aopt,:].T
     T = dot(a, W)
@@ -166,8 +188,8 @@ def bridge(a, b, aopt, scale='scores', mode='normal', r=0):
         return {'T':T, 'W':W}
 
     U = dot(g0, K) #fixme check this 
-    Q = dot(b.T, dot(T, inv(dot(T.T,T)) ))
-    B = zeros((aopt, n, l))
+    Q = dot(b.T, dot(T, inv(dot(T.T, T)) ))
+    B = zeros((aopt, n, l), dtype='f')
     for i in range(aopt):
         B[i] = dot(W[:,:i+1], Q[:,:i+1].T)
     # leverages
@@ -198,3 +220,6 @@ def bridge(a, b, aopt, scale='scores', mode='normal', r=0):
         
     return {'B':B, 'W':W, 'T':T, 'Q':Q, 'E':E, 'F':F, 'U':U, 'P':W}
     
+
+def m_shape(array):
+    return matrix(array).shape