whitespace

fluents/lib/validation.py

@@ -10,11 +10,15 @@ from engines import w_simpls,pls,bridge,pca
 from cx_utils import m_shape
 
 def w_pls_cv_val(X, Y, amax, n_blocks=None, algo='simpls'):
-    """Returns and  RMSEP for pls tailored for wide X.
+    """Returns rmsep and aopt for pls tailored for wide X.
+
+
+    comments:
+             -- X, Y inputs need to be centered (fixme: check)
+
     """
     k, l = m_shape(Y)
     PRESS = zeros((l, amax+1), dtype='f')
-    # X,Y are centered0
     if n_blocks==None:
         n_blocks = Y.shape[0]
     XXt = dot(X, X.T)
@@ -29,7 +33,7 @@ def w_pls_cv_val(X, Y, amax, n_blocks=None, algo='simpls'):
             That = dot(Doi, dot(U, inv(triu(dot(H.T,U))) ))
         else:
             raise NotImplementedError
-        #Yhat = empty((amax, k, l),dtype='<f8')
+        
         Yhat = []
         for j in range(l):
             TQ = dot(That, triu(dot(Q[j,:][:,newaxis], ones((1,amax)))) )
@@ -41,15 +45,21 @@ def w_pls_cv_val(X, Y, amax, n_blocks=None, algo='simpls'):
     aopt = find_aopt_from_sep(rmsep)
     return rmsep, aopt
 
-def pls_val(X, Y, amax=2, n_blocks=10,algo='pls'):
+def pls_val(X, Y, amax=2, n_blocks=10, algo='pls', metric=None):
     """ Validation results of pls model. 
+
+
+
+    comments:
+             -- X, Y inputs need to be centered (fixme: check)
+
+
     """    
     k, l = m_shape(Y)
     PRESS = zeros((l, amax+1), dtype='<f8')
     EE = zeros((amax, k, l), dtype='<f8')
     Yhat = zeros((amax, k, l), dtype='<f8')
-    # X,Y are centered
-    V = pls_gen(X, Y, n_blocks=n_blocks, center=True, index_out=True)
+    V = pls_gen(X, Y, n_blocks=n_blocks, center=True, index_out=True, metric=metric)
     for Xin, Xout, Yin, Yout, out in V:
         ym = -sum(Yout,0)[newaxis]/Yin.shape[0]
         Yin = (Yin - ym)
@@ -73,8 +83,14 @@ def pls_val(X, Y, amax=2, n_blocks=10,algo='pls'):
 
 def pca_alter_val(a, amax, n_sets=10, method='diag'):
     """Pca validation by altering elements in X.
+
+    
+
+
+    comments:
+             -- may do all jk estimates in this loop
     """
-    # todo: it is just as easy to do jk-estimates her as well
+    
     V = diag_pert(a, n_sets, center=True, index_out=True)
     sep = empty((n_sets, amax), dtype='f')
     for i, (xi, ind) in enumerate(V):
@@ -100,7 +116,9 @@ def pca_cv_val(a, amax, n_sets):
     output:
           -- sep, (amax x m x n), squared error of prediction (press)
           -- aopt, guestimated optimal number of components
+
     """
+
     m, n = a.shape
     E = empty((amax, m, n), dtype='f')
     xtot = (a**2).sum() # this needs centering
@@ -117,9 +135,10 @@ def pca_cv_val(a, amax, n_sets):
         sep.append(E[a].sum()/xtot)
     sep = array(sep)
     aopt = find_aopt_from_sep(sep)
+
     return sep, aopt
 
-def pls_jkW(a, b, amax, n_blocks=None, algo='pls', use_pack=False, center=True):
+def pls_jkW(a, b, amax, n_blocks=None, algo='pls', use_pack=True, center=True, metric=None):
     """ Returns CV-segments of paramter W for wide X.
     
     todo: add support for T,Q and B
@@ -129,26 +148,28 @@ def pls_jkW(a, b, amax, n_blocks=None, algo='pls', use_pack=False, center=True):
 
     Wcv = empty((n_blocks, a.shape[1], amax), dtype='f')
 
-    if use_pack:
+    if use_pack and metric==None:
         u, s, inflater = svd(a, full_matrices=0)
         a = u*s
     
-    V = pls_gen(a, b, n_blocks=n_blocks, center=center)
+    V = pls_gen(a, b, n_blocks=n_blocks, center=center, metric=metric)
     for nn,(a_in, a_out, b_in, b_out) in enumerate(V):
         if algo=='pls':
             dat = pls(a_in, b_in, amax, 'loads', 'fast')
+
         elif algo=='bridge':
             dat = bridge(a_in, b_in, amax, 'loads', 'fast')
+
         W = dat['W']
 
-        if use_pack:
+        if use_pack and metric==None:
             W = dot(inflater.T, W)
 
         Wcv[nn,:,:] = W
         
     return Wcv
 
-def pca_jkP(a, aopt, n_blocks=None):
+def pca_jkP(a, aopt, n_blocks=None, metric=None):
     """Returns loading from PCA on CV-segments.
     
     input:
@@ -178,9 +199,11 @@ def pca_jkP(a, aopt, n_blocks=None):
         
     return PP
 
+
 def find_aopt_from_sep(sep, method='75perc'):
     """Returns an estimate of optimal number of components from rmsecv.
     """
+
     if method=='vanilla':
         # min rmsep
         rmsecv = sqrt(sep.mean(0))