Merge pull request #15 from tensorflow/gfile

google3 updates

cav_test.py

@@ -60,7 +60,7 @@ def setUp(self):
     if os.path.exists(self.cav_dir):
       shutil.rmtree(self.cav_dir)
     os.mkdir(self.cav_dir)
-    with open(self.save_path, 'w') as pkl_file:
+    with tf.gfile.Open(self.save_path, 'w') as pkl_file:
       pickle.dump({
           'concepts': self.concepts,
           'bottleneck': self.bottleneck,

model.py

@@ -91,6 +91,21 @@ def adjust_prediction(self, pred_t):
     """
     return pred_t
 
+  def reshape_activations(self, layer_acts):
+    """Reshapes layer activations as needed to feed through the model network.
+
+    Override this for models that require reshaping of the activations for use
+    in TCAV.
+
+    Args:
+      layer_acts: Activations as returned by run_imgs.
+
+    Returns:
+      Activations in model-dependent form; the default is a squeezed array (i.e.
+      at most one dimensions of size 1).
+    """
+    return np.asarray(layer_acts).squeeze()
+
   @abstractmethod
   def label_to_id(self, label):
     """Convert label (string) to index in the logit layer (id)."""

tcav_helpers.py

@@ -41,27 +41,27 @@ def make_key(a_dict, key):
 
 def save_np_array(array, path):
   """Save an array as numpy array (loading time is better than pickling."""
-  with open(path, 'w') as f:
+  with tf.gfile.Open(path, 'w') as f:
     np.save(f, array, allow_pickle=False)
 
 
 def read_np_array(path):
   """Read a saved numpy array and return."""
-  with open(path) as f:
+  with tf.gfile.Open(path) as f:
     data = np.load(f)
   return data
 
 
 def read_file(path):
   """Read a file in path ."""
-  with open(path, 'r') as f:
+  with tf.gfile.Open(path, 'r') as f:
     data = f.read()
   return data
 
 
 def write_file(data, path, mode='w'):
   """Wrtie data to path to cns."""
-  with open(path, mode) as f:
+  with tf.gfile.Open(path, mode) as f:
     if mode == 'a':
       f.write('\n')
     f.write(data)
@@ -153,17 +153,14 @@ def load_images_from_files(filenames, max_imgs=500, return_filenames=False,
   else:
     return np.array(imgs)
 
-""" highe level overview.
+""" high level overview.
 get_acts_from_images: run images on a model and return activations.
 get_imgs_and_acts_save: loads images from image path and
                          calls get_acts_from_images to get images
                          and save them.
-
 """
 
 
-
-
 def get_acts_from_images(imgs, model, bottleneck_name):
   """Run images in the model to get the activations.
 
@@ -175,21 +172,22 @@ def get_acts_from_images(imgs, model, bottleneck_name):
   Returns:
     numpy array of activations.
   """
-  return np.asarray(model.run_imgs(imgs, bottleneck_name)).squeeze()
+  img_acts = model.run_imgs(imgs, bottleneck_name)
+  return model.reshape_activations(img_acts)
 
 
 def get_imgs_and_acts_save(model, bottleneck_name, img_paths, acts_path,
                            img_shape, max_images=500):
   """Get images from files, process acts and saves.
-  
+
   Args:
     model: a model instance
     bottleneck_name: name of the bottleneck that activations are from
     img_paths: where image lives
     acts_path: where to store activations
     img_shape: shape of the image.
     max_images: maximum number of images to save to acts_path
-  
+
   Returns:
     success or not.
   """
@@ -198,7 +196,7 @@ def get_imgs_and_acts_save(model, bottleneck_name, img_paths, acts_path,
   tf.logging.info('got %s imgs' % (len(imgs)))
   acts = get_acts_from_images(imgs, model, bottleneck_name)
   tf.logging.info('Writing acts to {}'.format(acts_path))
-  with open(acts_path, 'w') as f:
+  with tf.gfile.Open(acts_path, 'w') as f:
     np.save(f, acts, allow_pickle=False)
   del acts
   del imgs
@@ -251,7 +249,7 @@ def process_and_load_activations(model, bottleneck_names, concepts,
                                max_images=max_images)
 
       if bottleneck_name not in acts[concept].keys():
-        with open(acts_path) as f:
+        with tf.gfile.Open(acts_path) as f:
           acts[concept][bottleneck_name] = np.load(f).squeeze()
           tf.logging.info('Loaded {} shape {}'.format(
               acts_path,