Minor layer docstring fixes

lasagne/layers/dense.py

@@ -123,7 +123,7 @@ class NINLayer(Layer):
     W=lasagne.init.GlorotUniform(), b=lasagne.init.Constant(0.),
     nonlinearity=lasagne.nonlinearities.rectify, **kwargs)
 
-    Network-in-network layer [1]_.
+    Network-in-network layer.
     Like DenseLayer, but broadcasting across all trailing dimensions beyond the
     2nd.  This results in a convolution operation with filter size 1 on all
     trailing dimensions.  Any number of trailing dimensions is supported,

lasagne/layers/dnn.py

@@ -105,6 +105,12 @@ def get_output_for(self, input, **kwargs):
 
 
 class MaxPool2DDNNLayer(Pool2DDNNLayer):  # for consistency
+    """
+    2D max-pooling layer
+
+    Subclass of :class:`Pool2DDNNLayer` fixing ``mode='max'``, provided for
+    compatibility to other ``MaxPool2DLayer`` classes.
+    """
     def __init__(self, incoming, pool_size, stride=None,
                  pad=(0, 0), **kwargs):
         super(MaxPool2DDNNLayer, self).__init__(incoming, pool_size, stride,

lasagne/layers/noise.py

@@ -14,7 +14,7 @@
 
 
 class DropoutLayer(Layer):
-    """Dropout layer [1]_, [2]_
+    """Dropout layer
 
     Sets values to zero with probability p. See notes for disabling dropout
     during testing.
@@ -32,7 +32,7 @@ class DropoutLayer(Layer):
     Notes
     -----
     The dropout layer is a regularizer that randomly sets input values to
-    zero, see references for why this might improve generalization.
+    zero; see [1]_, [2]_ for why this might improve generalization.
     During training you should set deterministic to false and during
     testing you should set deterministic to true.
 
@@ -86,16 +86,16 @@ def get_output_for(self, input, deterministic=False, **kwargs):
 
 
 class GaussianNoiseLayer(Layer):
-    """Gaussian noise layer [1]_.
+    """Gaussian noise layer.
 
-    Add zero Gaussian noise with mean 0 and std sigma to the input
+    Add zero-mean Gaussian noise of given standard deviation to the input [1]_.
 
     Parameters
     ----------
     incoming : a :class:`Layer` instance or a tuple
             the layer feeding into this layer, or the expected input shape
     sigma : float or tensor scalar
-            Std of added Gaussian noise
+            Standard deviation of added Gaussian noise
 
     Notes
     -----

lasagne/layers/pool.py

@@ -364,7 +364,7 @@ def get_output_for(self, input, **kwargs):
 
 class FeatureWTALayer(Layer):
     """
-    'Winner Take' All layer
+    'Winner Take All' layer
 
     This layer performs 'Winner Take All' (WTA) across feature maps: zero out
     all but the maximal activation value within a region.