ENH: Added cross-validation to the module OnPLs.resampling.

Author: tomlof

OnPLS/__init__.py

@@ -13,8 +13,9 @@
 """
 from OnPLS import consts
 from OnPLS import estimators
+from OnPLS import resampling
 from OnPLS import utils
 
 __version__ = "0.0.1"
 
-__all__ = ["consts", "estimators", "utils"]
+__all__ = ["consts", "estimators", "resampling", "utils"]

OnPLS/estimators.py

@@ -27,7 +27,7 @@
 except ValueError:
     import OnPLS.consts as consts  # When run as a program.
 
-__all__ = ["BaseEstimator",
+__all__ = ["BaseEstimator", "BaseUniblock", "BaseTwoblock", "BaseMultiblock",
 
            "PCA", "nPLS", "OnPLS"]
 
@@ -133,7 +133,19 @@ def warn(self, *strs):
                 warnings.warn(warning)
 
 
-class PCA(BaseEstimator):
+class BaseUniblock(object):
+    pass
+
+
+class BaseTwoblock(object):
+    pass
+
+
+class BaseMultiblock(object):
+    pass
+
+
+class PCA(BaseUniblock, BaseEstimator):
     """A NIPALS implementation of principal components analysis.
 
     Parameters
@@ -358,7 +370,7 @@ def _deflate(self, X, p, t):
         return X - np.dot(t, p.T)
 
 
-class nPLS(BaseEstimator):
+class nPLS(BaseMultiblock, BaseEstimator):
     """The nPLS method for multiblock data analysis.
 
     Parameters
@@ -684,7 +696,7 @@ def generateA(self, X, psd=True):
         return A
 
 
-class OnPLS(BaseEstimator):
+class OnPLS(BaseMultiblock, BaseEstimator):
     """The OnPLS method for multiblock data analysis.
 
     Parameters
@@ -1033,7 +1045,7 @@ def predict(self, X, which=[], return_scores=False):
             if woi is None:
                 continue
             poi = self.Po[i]
-            ki = self.Wo[0].shape[1]
+            ki = self.Wo[i].shape[1]
             for k in range(ki):
                 woik = woi[:, [k]]
                 poik = poi[:, [k]]
@@ -1075,8 +1087,8 @@ def predict(self, X, which=[], return_scores=False):
                         Ti = T[i]
                         Tik = Ti[:, [k]]
                         Tiks.append(Tik)
+                Tiks = np.hstack(Tiks)
                 if len(Tiks) > 0:
-                    Tiks = np.hstack(Tiks)
                     beta = np.dot(np.linalg.pinv(Tiks), Tw[:, [k]])
                     Thatwk = np.dot(Tiks, beta)
                     Xhatw = Xhatw + np.dot(Thatwk, Pw.T)

OnPLS/resampling.py

@@ -0,0 +1,51 @@
+# -*- coding: utf-8 -*-
+"""
+The :mod:`OnPLS.resampling` module contains resampling methods that can be used
+to determine statistics from an estimator.
+
+Created on Wed Nov 23 21:02:03 2016
+
+Copyright (c) 2016, Tommy Löfstedt. All rights reserved.
+
+@author:  Tommy Löfstedt
+@email:   [email protected]
+@license: BSD 3-clause.
+"""
+import numpy as np
+
+import OnPLS.estimators as estimators
+
+__all__ = ["cross_validation"]
+
+
+def cross_validation(estimator, X, cv_rounds=7, random_state=None):
+
+    if isinstance(estimator, estimators.BaseUniblock):
+        if isinstance(X, np.ndarray):
+            X = [X]
+
+    n = len(X)
+
+    N = X[0].shape[0]
+
+    cv_rounds = min(max(1, N), int(cv_rounds))
+
+    scores = []
+    for k in range(cv_rounds):
+
+        test_samples = list(range(k, N, cv_rounds))
+        train_samples = list(set(range(0, N)).difference(test_samples))
+
+        Xtest = [0] * n
+        Xtrain = [0] * n
+        for i in range(n):
+            Xi = X[i]
+            Xtest[i] = Xi[test_samples, :]
+            Xtrain[i] = Xi[train_samples, :]
+
+        estimator.fit(Xtrain)
+
+        score = estimator.score(Xtest)
+        scores.append(score)
+
+    return scores

README.md

@@ -84,4 +84,6 @@ Xhat = onpls.predict([X1, X2, X3])
 
 # Compute prediction score
 score = onpls.score([X1, X2, X3])
+
+cv_scores = OnPLS.resampling.cross_validation(onpls, [X1, X2, X3], cv_rounds=4)
 ```