See ticket #417: http://www.pylearn.org/theano/trac/ticket/417

removed type mismatch warning from gradient propagation algorithm.  It was not a good idea. Its ok for a gradient to have a different type from the original thing.

Summary of changes:
 - Added implementations of functools.partial, defaultdict, any, all
 - Re-wrote relative imports to be absolute imports
 - Re-wrote conditional expressions to normal if/else expressions