Edit the backpropagation functions in cnn_model

cnn_model.py

@@ -66,7 +66,6 @@ def back_relu(layer, deriv):
     return np.multiply(layer, deriv)
 
 
-# TODO get the backpropagation for pooling to work
 def back_pool(layer, next_layer, deriv, size, stride=None):
     if stride is None:
         stride = size
@@ -82,8 +81,6 @@ def back_pool(layer, next_layer, deriv, size, stride=None):
 
 def back_conv(layer, deriv, size, stride=[1, 1]):
     derivW = np.resize(deriv, (len(deriv), len(deriv[0]) * len(deriv[0][0])))
-    print(deriv.shape)
-    print(derivW.shape)
     dw = np.zeros((len(layer), len(deriv), size[0] * size[1]))
     for i in range(len(layer)):
         for j in range(len(deriv)):

testing.py

@@ -3,7 +3,7 @@
 
 layer = np.random.random_integers(-10, 10, size=(1, 100, 100))
 filter_size = [2, 3]
-stride = [2, 3]
+stride = [1, 1]
 # first dim is the current layer size, second dim is the next layer size, and third dim is the total filter size
 filters = [np.random.random_integers(-5, 5, size=(3, 2, 6)), np.random.random_integers(-5, 5, size=(2, 5, 6))]
 
@@ -17,7 +17,7 @@
 
 dm2 = back_relu(m2, 1)
 dc2 = back_pool(c2, m2, dm2, [2, 2])
-dr1 = back_conv(r1, dc2, filter_size, stride)
+dr1 = back_conv(r1, dc2, filter_size, stride=stride)
 
 print(dr1)
 '''