.

data/cifar10.py

@@ -3,7 +3,7 @@
 import os
 
 # from http://cs.nyu.edu/~roweis/data.html
-path = os.path.dirname(__file__)+'/'
+path = os.environ['ML_DATA_PATH']+'/cifar10/'
 
 # Load the original images into numpy arrays
 def load_numpy():

data/freyface.py

@@ -3,7 +3,7 @@
 import os
 
 # from http://cs.nyu.edu/~roweis/data.html
-path = os.path.dirname(__file__)+'/frey_rawface.mat'
+path = os.environ['ML_DATA_PATH']+'/freyface/frey_rawface.mat'
 
 def load_numpy():
 	return scipy.io.loadmat(path)['ff'].T/256.

data/lfw.py

@@ -1,16 +1,18 @@
 import numpy as np
 import os
 
-path = os.path.dirname(__file__)+'/'
+path = os.environ['ML_DATA_PATH']+'/lfw/'
 
 # shape: (13233, 31, 23, 3)
 def load_lfw_small():
-	data = np.load(path+'lfw_31x23.npy').swapaxes(1,3).swapaxes(2,3).reshape((-1, 31*23*3))
-	data = data.astype('float16')/256.
-	return data, (31, 23)
+    data = np.load(path+'lfw_31x23.npy').swapaxes(1,3).swapaxes(2,3).reshape((-1, 31*23*3))
+    data = data.astype('float16')/256.
+    labels = np.load(path+'lfw_labels.npy')
+    return data, labels, (31, 23)
 
 def load_lfw_big():
-	data = np.load(path+'lfw_62x47.npy').swapaxes(1,3).swapaxes(2,3).reshape((-1, 62*47*3))
-	data = data.astype('float16')/256.
-	return data, (62, 47)
+    data = np.load(path+'lfw_62x47.npy').swapaxes(1,3).swapaxes(2,3).reshape((-1, 62*47*3))
+    data = data.astype('float16')/256.
+    labels = np.load(path+'lfw_labels.npy')
+    return data, labels, (62, 47)
 

data/lfw_generate.py

@@ -4,6 +4,7 @@
 # 'big' images (scale factor 0.5)
 lfw_people = fetch_lfw_people(color=True)
 np.save('lfw_62x47', lfw_people.images.astype('uint8'))
+np.save('lfw_labels', lfw_people.target)
 
 # 'small' images (scale factor 0.25)
 lfw_people2 = fetch_lfw_people(color=True, resize=0.25)

data/svhn.py

@@ -0,0 +1,29 @@
+import numpy as np
+import scipy.io
+import os
+
+# from http://cs.nyu.edu/~roweis/data.html
+path = os.environ['ML_DATA_PATH']
+
+def load_numpy(toFloat=True, binarize_y=False):
+    train = scipy.io.loadmat(path+'/svhn/train_32x32.mat')
+    train_x = train['X'].swapaxes(0,1).T.reshape((train['X'].shape[3], -1)).T
+    train_y = train['y'].reshape((-1)) - 1
+    test = scipy.io.loadmat(path+'/svhn/test_32x32.mat')
+    test_x = test['X'].swapaxes(0,1).T.reshape((test['X'].shape[3], -1)).T
+    test_y = test['y'].reshape((-1)) - 1
+    if toFloat:
+        train_x = train_x.astype('float16')/256.
+        test_x = test_x.astype('float16')/256.
+    if binarize_y:
+        train_y = binarize_labels(train_y)
+        test_y = binarize_labels(test_y)
+
+    return train_x, train_y, test_x, test_y 
+
+# Converts integer labels to binarized labels (1-of-K coding)
+def binarize_labels(y, n_classes=10):
+    new_y = np.zeros((n_classes, y.shape[0]))
+    for i in range(y.shape[0]):
+        new_y[y[i], i] = 1
+    return new_y