Remove the old way of silencing warnings.

theano/gof/tests/test_opt.py

@@ -386,8 +386,6 @@ class TestEquilibrium(object):
         g = Env([x, y, z], [e])
         print 'before', g
         # display pesky warnings along with stdout
-        oldstderr = sys.stderr
-        sys.stderr = sys.stdout
         # also silence logger for 'theano.gof.opt'
         _logger = logging.getLogger('theano.gof.opt')
         oldlevel = _logger.getEffectiveLevel()
@@ -401,7 +399,6 @@ class TestEquilibrium(object):
                 max_use_ratio = 1. / len(g.nodes)) # each opt can only be applied once
             opt.optimize(g)
         finally:
-            sys.stderr = oldstderr
             _logger.setLevel(oldlevel)
         print 'after', g
         assert str(g) == '[Op4(x, y)]'

theano/tensor/nnet/nnet.py

@@ -353,6 +353,9 @@ def local_softmax_with_bias(node):
             vectors = []
             non_vectors = []
             for x_in in x.owner.inputs:
+                print 'x_in =', x_in
+                print 'x_in.type =', x_in.type
+                print 'x_in.broadcastable =', x_in.broadcastable
                 if list(x_in.type.broadcastable) == [True, False]:
                     # print isinstance(x_in.owner.op, tensor.DimShuffle)
                     #since specialization comes relatively late in optimization, 
@@ -1239,7 +1242,7 @@ def categorical_crossentropy(coding_dist, true_dist):
 
     """
     if true_dist.ndim == coding_dist.ndim:
-        return -theano.sum(true_dist * log(coding_dist), axis=coding_dist.ndim-1)
+        return -tensor.sum(true_dist * tensor.log(coding_dist), axis=coding_dist.ndim-1)
     elif true_dist.ndim == coding_dist.ndim - 1:
         return crossentropy_categorical_1hot(coding_dist, true_dist)
     else: