pep8 fixes

.travis.yml

@@ -15,10 +15,10 @@ install:
   - pip install numpy matplotlib colorlog pep8 pytest protobuf catkin_pkg
   - pip install --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.10.0-cp27-none-linux_x86_64.whl
 script:
-  - python setup.py develop 
+  - python setup.py install 
   - pytest src/tensorflow_node/tests
   - pep8 --ignore=E501 .
 # Disable daemon branch for now...
-#branches:
-#  except:
-#    - daemon
+branches:
+  except:
+    - development

src/tensorflow_node/__init__.py

@@ -5,5 +5,3 @@
 from .nodes import AutoEncoderNode
 from .nodes import StackedAutoEncoderNode
 from .architectures import *
-
-#from ..destin.msg import DestinNodeState

src/tensorflow_node/architectures/architecture.py

@@ -7,6 +7,7 @@
 
 from tensorflow_node.nodes import *
 
+
 class NetworkArchitecture(object):
     __metaclass__ = abc.ABCMeta
 
@@ -15,10 +16,9 @@ def __init__(self):
         self.nodes = []
         pass
 
-
     def str_to_class(self, str):
         return getattr(sys.modules[__name__], str)
 
     def create_node(self, session, node_type, node_params):
         node_class = self.str_to_class(node_type)
-        return node_class(session, **node_params)
+        return node_class(session, **node_params)

src/tensorflow_node/architectures/destin.py

@@ -9,13 +9,14 @@
 from tensorflow_node.nodes import *
 from tensorflow_node.architectures import NetworkArchitecture
 
+
 class DestinArchitecture(NetworkArchitecture):
 
-    def __init__(self, session, inputlayer, node_type, node_params, receptive_field=[14,14], stride=[7,7]):
-        # TODO Assertions: 
+    def __init__(self, session, inputlayer, node_type, node_params, receptive_field=[14, 14], stride=[7, 7]):
+        # TODO Assertions:
         #   - inputlayer size and receptive field / stride fit together...
         #   - ...?
-    
+
         self.nodes = []
         self.train_op = []
 
@@ -25,33 +26,25 @@ def destin_node(level, number_of_layers, x_pos=0.0, y_pos=0.0):
             print " creating node @ level %i" % level
             node = self.create_node(session, node_type, node_params)
 
-            if (level<number_of_layers):
-                node.register_tensor(destin_node(level+1, number_of_layers, x_pos, y_pos))
-                node.register_tensor(destin_node(level+1, number_of_layers, x_pos+stride[0], y_pos))
-                node.register_tensor(destin_node(level+1, number_of_layers, x_pos, y_pos+stride[1]))
-                node.register_tensor(destin_node(level+1, number_of_layers, x_pos+stride[0], y_pos+stride[1]))
+            if (level < number_of_layers):
+                node.register_tensor(destin_node(level + 1, number_of_layers, x_pos, y_pos))
+                node.register_tensor(destin_node(level + 1, number_of_layers, x_pos + stride[0], y_pos))
+                node.register_tensor(destin_node(level + 1, number_of_layers, x_pos, y_pos + stride[1]))
+                node.register_tensor(destin_node(level + 1, number_of_layers, x_pos + stride[0], y_pos + stride[1]))
             else:
-                region = [int(np.round(x_pos)),int(np.round(y_pos)),receptive_field[0],receptive_field[1]]
+                region = [int(np.round(x_pos)), int(np.round(y_pos)), receptive_field[0], receptive_field[1]]
                 print "  registering region @ %i %i" % (x_pos, y_pos)
                 node.register_tensor(inputlayer.get_tensor_for_region(region))
 
             node.initialize_graph()
-            
+
             self.nodes.append(node)
             self.train_op.append(node.train_op)
-            
+
             return node.get_output_tensor()
 
         # calculate number of levels needed...
-        nr_of_layers = np.floor(np.log(np.power(inputlayer.output_size[0]/stride[0],2))/np.log(4))
-        
+        nr_of_layers = np.floor(np.log(np.power(inputlayer.output_size[0] / stride[0], 2)) / np.log(4))
+
         # create network
         destin_node(0, nr_of_layers)
-
-
-
-
-
-
-
-

src/tensorflow_node/architectures/handcoded_destin.py

@@ -7,6 +7,7 @@
 
 from tensorflow_node.architectures import NetworkArchitecture
 
+
 class HandcodedDestinArchitecture(NetworkArchitecture):
 
     def __init__(self, session, inputlayer, node_type, node_params):
@@ -17,13 +18,13 @@ def __init__(self, session, inputlayer, node_type, node_params):
 
         node_params["name"] = "bottom_b"
         ae_bottom_b = self.create_node(session, node_type, node_params)
-        
+
         node_params["name"] = "bottom_c"
         ae_bottom_c = self.create_node(session, node_type, node_params)
 
         node_params["name"] = "bottom_d"
         ae_bottom_d = self.create_node(session, node_type, node_params)
-        
+
         node_params["name"] = "top"
         ae_top = self.create_node(session, node_type, node_params)
 
@@ -41,5 +42,3 @@ def __init__(self, session, inputlayer, node_type, node_params):
 
         self.nodes = [ae_bottom_a, ae_bottom_b, ae_bottom_c, ae_bottom_d, ae_top]
         self.train_op = [ae_bottom_a.train_op, ae_bottom_b.train_op, ae_bottom_c.train_op, ae_bottom_d.train_op, ae_top.train_op]
-
-

src/tensorflow_node/input/opencv.py

@@ -20,7 +20,7 @@ def __init__(self, batch_size=1, output_size=[28, 28], input="", number_of_frame
         self.repeat = repeat
 
     def feed_to(self, feed_callback):
-        
+
         # TODO: there should be clearer distinction here, get these params via daemon.
         frames = rospy.get_param("tensorflow_node/inputlayer/params/number_of_frames")
         repeat = rospy.get_param("tensorflow_node/inputlayer/params/repeat")

src/tensorflow_node/input/ros.py

@@ -6,9 +6,9 @@
 from skimage import transform
 
 from tensorflow_node.input import InputLayer
-
 from sensor_msgs.msg import Image
 
+
 class ROSInputLayer(InputLayer):
     """
     Contains ROS to feed in images and video feeds to TF.
@@ -22,8 +22,8 @@ def callback(ros_data):
             np_arr = np.fromstring(ros_data.data, np.uint8).reshape(ros_data.width, ros_data.height, 3)
 
             # Grayscale conversion
-            channels = np_arr.swapaxes(0,2)
-            gray = (channels[0] + channels[1] + channels[2]) / 3 # could to different weights per channel here
+            channels = np_arr.swapaxes(0, 2)
+            gray = (channels[0] + channels[1] + channels[2]) / 3  # could to different weights per channel here
 
             # Resize to normalized input layer size
             resized = transform.resize(gray, [self.output_size[0], self.output_size[1]]).reshape([self.output_size[0], self.output_size[1], 1])
@@ -43,7 +43,7 @@ def callback(ros_data):
 
                 rospy.loginfo("ROSInputLayer: Evaluated batch")
 
-        ## ROS subscribe...
+        # ROS subscribe...
         rospy.logwarn("Subscribing to topic " + self.input)
         topic_name = self.input
         rospy.Subscriber(topic_name, Image, callback)

src/tensorflow_node/nodes/autoencoder.py

@@ -30,10 +30,10 @@ def __init__(self,
                  lr=0.007):
 
         self.name = name
-        
-        if self.name=="ae":
+
+        if self.name == "ae":
             self.name = 'ae_%08x' % random.getrandbits(32)
-        
+
         self.session = session
 
         # this list is populated with register tensor function

src/tensorflow_node/utils/summary_writer.py

@@ -9,7 +9,7 @@
 
 import rospy
 
-# Singleton
+
 class SummaryWriter(object):
     _instance = None
 
@@ -26,7 +26,7 @@ def __init__(self):
             self.writer = tf.train.SummaryWriter(self.directory)
 
     def get_output_folder(self, path):
-        #output_path = pjoin(os.getcwd(), 'output', path)
+        # output_path = pjoin(os.getcwd(), 'output', path)
         output_path = rospy.get_param("tensorflow_node/publishing/summary_folder")
         if not os.path.exists(output_path):
             os.makedirs(output_path)