diff --git a/examples/cifar10_nasnet.py b/examples/cifar10_nasnet.py
index 11aba03..56c75ee 100644
--- a/examples/cifar10_nasnet.py
+++ b/examples/cifar10_nasnet.py
@@ -84,6 +84,9 @@ else:
     datagen.fit(X_train)
 
     # wrap the ImageDataGenerator to yield two label batches [y, y] for each input batch X
+    # When training a NASNet model, we have to use its auxilary training head
+    # Therefore the model is technically a 1 input - 2 output model, and requires
+    # the label to be duplicated for the auxilary head
     def image_data_generator_wrapper(image_datagenerator, batch_size):
         iterator = datagen.flow(X_train, Y_train, batch_size=batch_size)
 
diff --git a/keras_contrib/applications/nasnet.py b/keras_contrib/applications/nasnet.py
index bab9e9f..12895a6 100644
--- a/keras_contrib/applications/nasnet.py
+++ b/keras_contrib/applications/nasnet.py
@@ -248,7 +248,7 @@ def NASNet(input_shape=None,
 
     # load weights (when available)
     if weights is not None:
-        warnings.warn('Weights of NASNet models have not been ported yet for Keras.')
+        warnings.warn('Weights of NASNet models have not yet been ported to Keras')
 
     if old_data_format:
         K.set_image_data_format(old_data_format)
@@ -684,6 +684,10 @@ def _reduction_A(ip, p, filters, weight_decay=5e-5, id=None):
 def _add_auxilary_head(x, classes, weight_decay):
     '''Adds an auxilary head for training the model
 
+    From section A.7 "Training of ImageNet models" of the paper, all NASNet models are
+    trained using an auxilary classifier around 2/3 of the depth of the network, with
+    a loss weight of 0.4
+
     # Arguments
         x: input tensor
         classes: number of output classes