Reorder arguments of h_softmax

theano/tensor/nnet/nnet.py

@@ -2052,11 +2052,10 @@ def relu(x, alpha=0):
         return f1 * x + f2 * abs(x)
 
 
-def h_softmax(x, batch_size, n_outputs, W1, b1, W2, b2,
-              n_classes, n_outputs_per_class, target=None):
+def h_softmax(x, batch_size, n_outputs, n_classes, n_outputs_per_class,
+              W1, b1, W2, b2, target=None):
     """ Two-level hierarchical softmax.
 
-    Outputs are grouped in sqrt(n_outputs) classes.
     The architecture is composed of two softmax layers: the first predicts the
     class of the input x while the second predicts the output of the input x in
     the predicted class.

theano/tensor/nnet/tests/test_nnet.py

@@ -1489,12 +1489,12 @@ def test_h_softmax():
     y = tensor.ivector('y')
 
     # This only computes the output corresponding to the target
-    y_hat_tg = h_softmax(x, batch_size, output_size, W1, b1, W2, b2,
-                         h_softmax_level1_size, h_softmax_level2_size, y)
+    y_hat_tg = h_softmax(x, batch_size, output_size, h_softmax_level1_size,
+                         h_softmax_level2_size, W1, b1, W2, b2, y)
 
     # This computes all the outputs
-    y_hat_all = h_softmax(x, batch_size, output_size, W1, b1, W2, b2,
-                          h_softmax_level1_size, h_softmax_level2_size)
+    y_hat_all = h_softmax(x, batch_size, output_size, h_softmax_level1_size,
+                          h_softmax_level2_size, W1, b1, W2, b2)
 
     #############
     # Compile functions