making it serializable

Author: aflag

break_captcha

@@ -41,7 +41,7 @@ def main():
                 else:
                     with open(img_path) as f_image:
                         image = Image.open(f_image).convert('L')
-                print model.decode_image(image)
+                print model.predict(image)
         except EOFError:
             pass
 

features.py

@@ -38,21 +38,29 @@ def __call__(self, arg):
             extractor(image, image_features)
         return image_features
 
-def _border_detection(digit):
-    digit.image = digit.image.filter(ImageFilter.FIND_EDGES)
-    digit.pix = digit.image.load()
-    return digit
+class border(object):
+    def __init__(self, callback):
+        self.callback = callback
 
-def border(callback):
-    return lambda digit,features: callback(_border_detection(digit), features, prefix='border-')
+    def __border_detection(self, digit):
+        digit.image = digit.image.filter(ImageFilter.FIND_EDGES)
+        digit.pix = digit.image.load()
+        return digit
 
-def _scale_down(digit):
-    digit.image = digit.image.resize((16,16), Image.BICUBIC)
-    digit.pix = digit.image.load()
-    return digit
+    def __call__(self, digit, features):
+        return callback(self.__border_detection(digit), features, prefix='border-')
 
-def scale_image_down(callback):
-    return lambda digit,features: callback(_scale_down(digit), features, prefix='scaled-')
+class scale_image_down(object):
+    def __init__(self, callback):
+        self.callback = callback
+
+    def __scale_down(self, digit):
+        digit.image = digit.image.resize((16,16), Image.BICUBIC)
+        digit.pix = digit.image.load()
+        return digit
+
+    def __call__(self, digit, features):
+        return self.callback(self.__scale_down(digit), features, prefix='scaled-')
 
 def is_white(color):
     return color > 230

models.py

@@ -27,8 +27,6 @@
 from sklearn import ensemble
 from sklearn.feature_extraction import DictVectorizer
 
-class ModelUnavailable(Exception):
-    pass
 
 ALL_EXTRACTORS = [
         x_histogram,
@@ -69,23 +67,26 @@ def fit(self, x, y):
         train_array = self.vectorizer.fit_transform(digits).toarray()
         self.engine.fit(train_array, labels)
 
+    def __make_prediction(self, image):
+        separator = DigitSeparator(image) 
+        features = map(self.feature_extractor, separator.get_digits())
+        digits = self.vectorizer.transform(features).toarray()
+        labels = self.engine.predict(digits)
+        return ''.join(map(lambda x: '%d'%x, labels))
+
     def predict(self, x):
-        prediction = []
-        for image in x:
-           separator = DigitSeparator(image) 
-           features = map(self.feature_extractor, separator.get_digits())
-           digits = self.vectorizer.transform(features).toarray()
-           labels = self.engine.predict(digits)
-           prediction.append(''.join(map(lambda x: '%d'%x, labels)))
-        return prediction
+        if not hasattr(x, '__iter__'):
+            return self.__make_prediction(x)
+        else:
+            prediction = []
+            for image in x:
+                prediction.append(self.__make_prediction(image))
+            return prediction
 
     def score(self, data, labels):
         pred_labels = self.predict(data)
         matches = sum(map(lambda (x,y): x==y, zip(labels, pred_labels)))
         return float(matches)/len(labels)
 
-    def decode_image(self, image):
-        return self.predict([image])[0]
-
     def get_params(self, *args, **kwargs):
         return self.engine.get_params(*args, **kwargs)

train.py

@@ -39,7 +39,7 @@ def main():
         joblib.dump(model, sys.argv[2])
     else:
         t0 = time.time()
-        scores = cross_validation.cross_val_score(model, numpy.array(dataset[0], dtype=object), dataset[1], cv=50)
+        scores = cross_validation.cross_val_score(model, numpy.array(dataset[0], dtype=object), dataset[1], cv=5)
         print 'Accuracy: %0.2f (+/- %0.2f)' % (scores.mean(), scores.std()/2)
         print 'Validation time:', time.time() - t0