Changin the policies to Tensorflow version

[elmiraloo]

I was wondering if there is any specific reason behind using Lasagne and Theano in defining different policies. I can change the code to Tensorflow (seems to be straightforward) and in case of resource allocation working with one main library seems to be a better idea. Am I missing something? If there is no specific reason I can re-factor the code.