Cleaning before handover 1

Author: ocloarec2

.gitignore

@@ -58,3 +58,5 @@ docs/_build/
 
 # PyBuilder
 target/
+
+.DS_Store

pyPLS/nopls.py

@@ -71,8 +71,8 @@ def __init__(self, X, Y, scaling=0,
         self.K = kernels[kernel](self.X, **kwargs)
          # Correction of the kernel matrix
         if penalization:
-
-            self.K = diagonal_correction(self.K, np.mean(self.Y, axis=1))
+            self.K  = self.K - np.diag(np.diag(self.K)) * penalization
+            # self.K = diagonal_correction(self.K, np.mean(self.Y, axis=1))
 
         assert not np.isnan(self.K).any(), "Kernel calculation lead to missing values!"
 
@@ -116,7 +116,8 @@ def __init__(self, X, Y, scaling=0,
 
         if statistics:
             #self.Yhat = self.predict(self.X, preprocessing=False, kernel=kernel)
-            self.Yhat = self.T @ self.C.T
+            # self.Yhat = self.T @ self.C.T
+            self.Yhat = self.predict(self.X, preprocessing=False)
             self.R2Y, self.R2Ycol = self._calculateR2Y(self.Yhat)
             if kernel == "linear":
                 self.R2X = np.sum(np.square(self.T @ self.P.T))/self.SSX

pyPLS/pls.py

@@ -66,41 +66,41 @@ def __init__(self, X, Y, ncp=1, cvfold=None, scaling=0):
 
 
     def predict(self, Xnew, preprocessing=True, statistics=False, **kwargs):
-
-
+    
+    
         Xnew, nnew, pxnew = isValid(Xnew, forPrediction=True)
         if preprocessing:
             Xnew = (Xnew - self.Xbar)
             Xnew /= np.power(self.Xstd, self.scaling)
-
+    
         assert pxnew == self.px, "New observations do not have the same number of variables!!"
-
+    
         if statistics:
             That = Xnew @ self.W
             Xpred = That @ self.P.T
             Xres = Xnew - Xpred
             Xnew2 = np.square(Xres)
-
+    
             if np.isnan(Xnew2).any():
                 ssrx = np.nansum(Xnew2, axis=0)
             else:
                 ssrx = np.sum(Xnew2, axis=0)
             stats = {'That':That, 'ESS':ssrx}
-
-
+    
+    
         if self.B is not None:
             # Yhat = Xnew @ self.B
-
+    
             if self.missingValuesInX:
                 Yhat = nanmatprod(Xnew, self.B)
             else:
                 Yhat = Xnew @ self.B
-
+    
             if preprocessing:
                 Yhat = Yhat * np.power(self.Ystd, self.scaling) + self.Ybar
         else:
             Yhat = None
-
+    
         if statistics:
             return Yhat, stats
         else:

pyPLS/preprocessing.py

@@ -2,6 +2,7 @@
 
 
 def diagonal_correction(K, v):
+
     n = K.shape[0]
     correction = np.zeros((n, n))
     H = (np.eye(n) - np.ones((n, n), dtype=float)/n)

requirements.txt

@@ -0,0 +1,3 @@
+numpy
+scipy
+pandas

setup.py

@@ -2,12 +2,12 @@
 
 setup(
     name='pyPLS',
-    version='0.1.1',
+    version='0.2.0',
     packages=find_packages(),
     url='',
     license='MIT',
     author='Olivier Cloarec',
-    author_email='ocloarec@korrigan.co.uk',
+    author_email='ocloarec@mac.com',
     description='A package for Projection on Latent Structures',
     requires=['numpy', 'pandas', 'scipy']
 )

test.py

@@ -5,33 +5,25 @@
 
 
 def test_pls():
-    out = pyPLS.nopls(Xt, Yt[:, 0], cvfold=7, scaling=1., penalization=True)
-    print("noPLS1 R2 = ", out.R2Y, out.ncp)
-    print("noPLS1 Q2 = ", out.Q2Y, out.ncp)
+    #out = pyPLS.nopls(Xt, Yt[:, 0], cvfold=7, scaling=1., penalization=True)
+    # print("noPLS1 R2 = ", out.R2Y, out.ncp)
+    # print("noPLS1 Q2 = ", out.Q2Y, out.ncp)
     out = pyPLS.pls(Xt, Yt[:, 0], ncp=2, scaling=1., cvfold=7)
     print("PLS1 R2 = ", out.R2Y, out.ncp)
     print("PLS1 Q2 = ", out.Q2Y, out.ncp)
-    out = pyPLS.nopls(Xt, Yt[:, 0:2], cvfold=7, scaling=1., penalization=True)
-    print("noPLS2 R2 = ", out.R2Ycol[0], out.R2Ycol[1], out.ncp)
-    print("noPLS2 Q2 = ", out.Q2Ycol[0], out.Q2Ycol[1], out.ncp)
+    # out = pyPLS.nopls(Xt, Yt[:, 0:2], cvfold=7, scaling=1., penalization=True)
+    # print("noPLS2 R2 = ", out.R2Ycol[0], out.R2Ycol[1], out.ncp)
+    # print("noPLS2 Q2 = ", out.Q2Ycol[0], out.Q2Ycol[1], out.ncp)
     out = pyPLS.pls(Xt, Yt[:, 0:2], ncp=2, cvfold=7, scaling=1.)
     print("PLS2 R2 = ", out.R2Ycol[0], out.R2Ycol[1], out.ncp)
     print("PLS2 Q2 = ", out.Q2Ycol[0], out.Q2Ycol[1], out.ncp)
 
 
-def test_kernel():
-    start_time = time.time()
-    K2 = pyPLS.linear(Xt, Y=Xt)
-    print("Linear Kernel --- %s seconds ---" % (time.time() - start_time))
-    start_time = time.time()
-    K2 = pyPLS.gaussian(Xt, sigma=3.0)
-    print("Gaussian Kernel --- %s seconds ---" % (time.time() - start_time))
-
 def test_pca():
     start_time = time.time()
-    # out = pyPLS.pca(Xt,4)
-    # print("PCA using svd on the kernel  --- %s seconds ---" % (time.time() - start_time))
-    # start_time = time.time()
+    out = pyPLS.pca(Xt,4)
+    print("PCA using svd on the kernel  --- %s seconds ---" % (time.time() - start_time))
+    start_time = time.time()
     out = pyPLS.pca(Xt, 4, method='nipals')
     print("PCA using nipals  --- %s seconds ---" % (time.time() - start_time))
     start_time = time.time()
@@ -42,20 +34,17 @@ def prediction():
 
     out = pyPLS.pls(Xt, Yt[:, 0], ncp=2, scaling=1., cvfold=7)
     Yhat, stats = out.predict(Xt, statistics=True)
-    print(stats['ESS'])
 
     out = pyPLS.pca(Xt, 3, scaling=1)
     That = out.predict(Xt)
-    print(That)
+
 
 if __name__ == '__main__':
     Xt, Z, Yt = pyPLS.simulateData(600, 2, 30, 5., signalToNoise=100.)
     # for i in np.arange(50):
     #     Xt[i,i] = np.NaN
     if "nopls" in sys.argv[1:]:
         test_pls()
-    if "kernel" in sys.argv[1:]:
-        test_kernel()
     if "pca" in sys.argv[1:]:
         test_pca()
     if "prediction" in sys.argv[1:]: