Added float64 warning to TensorVariable class.

theano/configdefaults.py

@@ -24,6 +24,11 @@ AddConfigVar('floatX',
              EnumStr('float64', 'float32', convert=floatX_convert,),
 )
 
+AddConfigVar('warn_float64',
+             "Warning for float64 output",
+            BoolParam(False)
+)
+
 AddConfigVar('cast_policy',
         "Rules for implicit type casting",
         EnumStr('custom', 'numpy+floatX',

theano/tensor/var.py

@@ -574,6 +574,14 @@ class _tensor_py_operators:
 class TensorVariable(_tensor_py_operators, Variable):
     """Subclass to add the tensor operators to the basic `Variable` class."""
 
+     def __init__(self, type)
+          if (config.warn_float64 and type = float64):
+                print '''Warning, you are instanciating parts of your graph 
+                    with the float64 type (possibly by accident).  
+                    This means that those parts will not run on the GPU.  
+                    If that is what you want, you can disable this warning by 
+                    setting warn_float64=False in your theano flags.'''
+
 TensorType.Variable = TensorVariable