Use correct face normalisation

1 files changed, 18 insertions, 12 deletions

diff --git a/train.py b/train.py
index 6ac963b..169163f 100755
--- a/train.py
+++ b/train.py
@@ -38,9 +38,9 @@ n_cases = 10
 n_pixels = 2576
 
 # subtract the normal face from each row of the face matrix
-def normalise_faces(average_face, faces):
+def normalise_faces(deviations_tr, average_face, faces):
     faces = np.subtract(faces, np.tile(average_face, (faces.shape[0],1)))
-    return np.divide(faces.T, np.std(faces.T, axis=0)).T
+    return np.divide(faces, deviations_tr)
 
 # Split data into training and testing sets
 def test_split(n_faces, raw_faces, split, seed):
@@ -82,16 +82,22 @@ def test_model(M, faces_train, faces_test, target_train, target_test, args):
 
     explained_variances = ()
 
+
+    distances = np.zeros(faces_test.shape[0])
+
     if args.pca or args.pca_r:
         # faces_pca containcts the principial components or the M most variant eigenvectors
         average_face = np.mean(faces_train, axis=0)
         #PLOTTING MEAN FACE
         #plt.imshow(average_face.reshape([46,56]).T, cmap = 'gist_gray')
         plt.show()
-        deviations_tr =  np.std(faces_train, axis=0)
-        deviations_tst = np.std(faces_train, axis=0)
-        faces_train = normalise_faces(average_face, faces_train)
-        faces_test = normalise_faces(average_face, faces_test)
+        if args.classifyalt:
+            deviations_tr = np.ones(n_pixels)
+        else:
+            deviations_tr =  np.std(faces_train, axis=0)
+        # deviations_tst = np.std(faces_test, axis=0)
+        faces_train = normalise_faces(deviations_tr, average_face, faces_train)
+        faces_test = normalise_faces(deviations_tr, average_face, faces_test)
         if (args.pca_r):
             e_vals, e_vecs = LA.eigh(np.dot(faces_train, faces_train.T))
             e_vecs = np.dot(faces_train.T, e_vecs)
@@ -107,13 +113,10 @@ def test_model(M, faces_train, faces_test, target_train, target_test, args):
         e_vals = np.flip(e_vals)[:M]
         e_vecs = np.fliplr(e_vecs).T[:M]
         deviations_tr = np.flip(deviations_tr)
-        deviations_tst = np.flip(deviations_tst)
+        # deviations_tst = np.flip(deviations_tst)
         faces_train = np.dot(faces_train, e_vecs.T)
         faces_test = np.dot(faces_test, e_vecs.T)
 
-        distances = np.zeros(faces_test.shape[0])
-
-
         rec_vecs = np.add(np.tile(average_face, (faces_test.shape[0], 1)), np.dot(faces_test, e_vecs) * deviations_tr)
         distances = LA.norm(raw_faces_test - rec_vecs, axis=1);
 
@@ -127,7 +130,7 @@ def test_model(M, faces_train, faces_test, target_train, target_test, args):
 
     if args.lda:
         if args.pca_r or (args.pca and M > n_training_faces - n_faces):
-            lda = LinearDiscriminantAnalysis(n_components=M, solver='eigen')
+            lda = LinearDiscriminantAnalysis(n_components=M, solver='svd')
         else:
             lda = LinearDiscriminantAnalysis(n_components=M, store_covariance='True')
 
@@ -136,7 +139,7 @@ def test_model(M, faces_train, faces_test, target_train, target_test, args):
         class_means = lda.means_
         e_vals = lda.explained_variance_ratio_
         scatter_matrix = lda.covariance_
-        print(LA.matrix_rank(scatter_matrix))
+        print("Rank of scatter:", LA.matrix_rank(scatter_matrix))
 
     if args.faces:
         if args.lda:
@@ -207,6 +210,9 @@ def main():
     parser.add_argument("-alt", "--classifyalt", help="Alternative method ON", action='store_true')
     args = parser.parse_args()
 
+    if args.lda and args.classifyalt:
+        sys.exit("LDA and Alt PCA can not be performed together")
+
     raw_faces = genfromtxt(args.data, delimiter=',')
     targets = np.repeat(np.arange(n_faces),n_cases)