Merge pull request #8 from stablum/master

Processing modes, including 'Randomly Binarized'

mnist/loader.py

@@ -1,12 +1,32 @@
 import os
 import struct
 from array import array
+import random
 
+_allowed_modes = (
+    # integer values in {0..255}
+    'vanilla',
+
+    # integer values in {0,1}
+    # values set at 1 (instead of 0) with probability p = orig/255
+    # as in Ruslan Salakhutdinov and Iain Murray's paper
+    # 'On The Quantitative Analysis of Deep Belief Network' (2008)
+    'randomly_binarized',
+
+    # integer values in {0,1}
+    # values set at 1 (instead of 0) if orig/255 > 0.5
+    'rounded_binarized'
+)
 
 class MNIST(object):
-    def __init__(self, path='.'):
+    def __init__(self, path='.', mode='vanilla'):
         self.path = path
 
+        assert mode in _allowed_modes, \
+            "selected mode '{}' not in {}".format(mode,_allowed_modes)
+
+        self._mode = mode
+
         self.test_img_fname = 't10k-images-idx3-ubyte'
         self.test_lbl_fname = 't10k-labels-idx1-ubyte'
 
@@ -19,11 +39,15 @@ def __init__(self, path='.'):
         self.train_images = []
         self.train_labels = []
 
+    @property # read only because set only once, via constructor
+    def mode(self):
+        return self._mode
+
     def load_testing(self):
         ims, labels = self.load(os.path.join(self.path, self.test_img_fname),
                                 os.path.join(self.path, self.test_lbl_fname))
 
-        self.test_images = ims
+        self.test_images = self.process(ims)
         self.test_labels = labels
 
         return ims, labels
@@ -32,11 +56,31 @@ def load_training(self):
         ims, labels = self.load(os.path.join(self.path, self.train_img_fname),
                                 os.path.join(self.path, self.train_lbl_fname))
 
-        self.train_images = ims
+        self.train_images = self.process(ims)
         self.train_labels = labels
 
         return ims, labels
 
+    def process(self, images):
+        if self.mode == 'vanilla':
+            pass # no processing, return them vanilla
+
+        elif self.mode == 'randomly_binarized':
+            for i in range(len(images)):
+                for j in range(len(images[i])):
+                    pixel = images[i][j]
+                    images[i][j] = int(random.random() <= pixel/255) # bool to 0/1
+
+        elif self.mode == 'rounded_binarized':
+            for i in range(len(images)):
+                for j in range(len(images[i])):
+                    pixel = images[i][j]
+                    images[i][j] = int(pixel/255 > 0.5) # bool to 0/1
+        else:
+            raise Exception("unknown mode '{}'".format(self.mode))
+
+        return images
+
     @classmethod
     def load(cls, path_img, path_lbl):
         with open(path_lbl, 'rb') as file: