temp

autokeras/bayesian.py

@@ -29,7 +29,7 @@ def skip_connection_distance(a, b):
         return 1.0
     len_a = abs(a[1] - a[0])
     len_b = abs(b[1] - b[0])
-    return abs(a[0] - b[0]) + abs(len_a - len_b)
+    return (abs(a[0] - b[0]) + abs(len_a - len_b)) / (max(a[0], b[0]) + max(len_a, len_b))
 
 
 def skip_connections_distance(list_a, list_b):

autokeras/classifier.py

@@ -159,7 +159,7 @@ def fit(self, x_train=None, y_train=None, csv_file_path=None, images_path=None,
         if not self.searcher:
             input_shape = x_train.shape[1:]
             n_classes = self.y_encoder.n_classes
-            searcher = self._get_searcher_class()(n_classes, input_shape, self.path, self.verbose)
+            searcher = self._get_searcher_class()(n_classes, input_shape, self.path, self.verbose, self.augment)
             self.save_searcher(searcher)
             self.searcher = True
 
@@ -273,5 +273,7 @@ class ImageClassifier(ClassifierBase):
     for the best configuration for the dataset.
     """
 
-    def __init__(self, verbose=True, searcher_type='bayesian', path=constant.DEFAULT_SAVE_PATH, resume=False):
+    def __init__(self, verbose=True, searcher_type='bayesian', path=constant.DEFAULT_SAVE_PATH, resume=False,
+                 augment=True):
         super().__init__(verbose, searcher_type, path, resume=resume)
+        self.augment = augment

autokeras/constant.py

@@ -26,4 +26,4 @@
 MAX_NO_IMPROVEMENT_NUM = 100
 MAX_BATCH_SIZE = 32
 LIMIT_MEMORY = False
-SEARCH_MAX_ITER = 10
+SEARCH_MAX_ITER = 30

autokeras/search.py

@@ -40,7 +40,7 @@ class Searcher:
         model_count: the id of model
     """
 
-    def __init__(self, n_classes, input_shape, path, verbose):
+    def __init__(self, n_classes, input_shape, path, verbose, augment=True):
         """Init Searcher class with n_classes, input_shape, path, verbose
 
         The Searcher will be loaded from file if it has been saved before.
@@ -53,6 +53,7 @@ def __init__(self, n_classes, input_shape, path, verbose):
         self.path = path
         self.model_count = 0
         self.descriptors = {}
+        self.augment = augment
 
     def search(self, x_train, y_train, x_test, y_test):
         """an search strategy that will be overridden by children classes"""
@@ -77,7 +78,7 @@ def get_best_model_id(self):
     def replace_model(self, model, model_id):
         model.save(os.path.join(self.path, str(model_id) + '.h5'))
 
-    def add_model(self, model, x_train, y_train, x_test, y_test, max_no_improve=constant.MAX_ITER_NUM):
+    def add_model(self, model, x_train, y_train, x_test, y_test, max_iter_num=constant.MAX_ITER_NUM):
         """add one model while will be trained to history list
 
         Returns:
@@ -90,7 +91,8 @@ def add_model(self, model, x_train, y_train, x_test, y_test, max_no_improve=cons
                      y_train,
                      x_test,
                      y_test,
-                     self.verbose).train_model(max_no_improvement_num=max_no_improve)
+                     self.verbose,
+                     augment=self.augment).train_model(max_iter_num=max_iter_num)
         loss, accuracy = model.evaluate(x_test, y_test, verbose=self.verbose)
         model.save(os.path.join(self.path, str(self.model_count) + '.h5'))
         plot_model(model, to_file=os.path.join(self.path, str(self.model_count) + '.png'), show_shapes=True)
@@ -180,8 +182,8 @@ def search(self, x_train, y_train, x_test, y_test):
 
 class BayesianSearcher(Searcher):
 
-    def __init__(self, n_classes, input_shape, path, verbose):
-        super().__init__(n_classes, input_shape, path, verbose)
+    def __init__(self, n_classes, input_shape, path, verbose, augment=True):
+        super().__init__(n_classes, input_shape, path, verbose, augment)
         self.gpr = IncrementalGaussianProcess()
         self.search_tree = SearchTree()
         self.init_search_queue = None
@@ -191,7 +193,7 @@ def __init__(self, n_classes, input_shape, path, verbose):
     def search(self, x_train, y_train, x_test, y_test):
         if not self.history:
             model = DefaultClassifierGenerator(self.n_classes, self.input_shape).generate()
-            history_item = self.add_model(model, x_train, y_train, x_test, y_test)
+            history_item = self.add_model(model, x_train, y_train, x_test, y_test, constant.SEARCH_MAX_ITER)
             self.search_tree.add_child(-1, history_item['model_id'])
 
             graph = Graph(model)

experiments/cifar10.py

@@ -22,13 +22,12 @@ def select_gpu():
 
 
 if __name__ == '__main__':
-    # select_gpu()
-    os.environ["CUDA_VISIBLE_DEVICES"] = str(3)
-    constant.LIMIT_MEMORY = True
+    select_gpu()
+    # constant.LIMIT_MEMORY = True
     (x_train, y_train), (x_test, y_test) = cifar10.load_data()
     X = np.concatenate((x_train, x_test))
     Y = np.concatenate((y_train, y_test))
-    clf = ImageClassifier(searcher_type='bayesian', path='~/temp_models', verbose=True)
+    clf = ImageClassifier(searcher_type='bayesian', path='/home/haifeng/cifar10', verbose=False)
 
     clf.fit(x_train, y_train, time_limit=12*60*60)
     # clf.final_fit(x_train, y_train)

experiments/fashion.py

@@ -22,15 +22,15 @@ def select_gpu():
 
 
 if __name__ == '__main__':
-    # select_gpu()
-    os.environ["CUDA_VISIBLE_DEVICES"] = str(3)
-    constant.LIMIT_MEMORY = True
+    select_gpu()
+    # os.environ["CUDA_VISIBLE_DEVICES"] = str(3)
+    # constant.LIMIT_MEMORY = True
     (x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
     x_train = x_train.reshape(x_train.shape + (1,))
     x_test = x_test.reshape(x_test.shape + (1,))
     X = np.concatenate((x_train, x_test))
     Y = np.concatenate((y_train, y_test))
-    clf = ImageClassifier(searcher_type='bayesian', path='~/temp_models', verbose=True)
+    clf = ImageClassifier(searcher_type='bayesian', path='/haifeng/home/fashion', verbose=False)
 
     clf.fit(x_train, y_train, time_limit=12*60*60)
     # clf.final_fit(x_train, y_train)

experiments/mnist.py

@@ -22,15 +22,15 @@ def select_gpu():
 
 
 if __name__ == '__main__':
-    # select_gpu()
-    os.environ["CUDA_VISIBLE_DEVICES"] = str(3)
-    constant.LIMIT_MEMORY = True
+    select_gpu()
+    # os.environ["CUDA_VISIBLE_DEVICES"] = str(3)
+    # constant.LIMIT_MEMORY = True
     (x_train, y_train), (x_test, y_test) = mnist.load_data()
     x_train = x_train.reshape(x_train.shape + (1,))
     x_test = x_test.reshape(x_test.shape + (1,))
     X = np.concatenate((x_train, x_test))
     Y = np.concatenate((y_train, y_test))
-    clf = ImageClassifier(searcher_type='bayesian', path='~/temp_models', verbose=True)
+    clf = ImageClassifier(searcher_type='bayesian', path='/home/haifeng/mnist', verbose=False)
 
     clf.fit(x_train, y_train, time_limit=12*60*60)
     # clf.final_fit(x_train, y_train)

tests/test_classifier.py

@@ -163,7 +163,7 @@ def test_cross_validate(_, _1):
     path = 'tests/resources/temp'
     clean_dir(path)
     clf = ImageClassifier(path=path, verbose=False)
-    train_x = np.random.rand(100, 25, 1)
+    train_x = np.random.rand(100, 25, 25, 1)
     train_y = np.random.randint(0, 5, 100)
     clf.fit(train_x, train_y)
     results = clf.cross_validate(train_x, train_y, 2)