Deterministic Theano Op - Gradient

[junpenglao]

@froeck

I am given a precomputed covariance matrix which contains floats, +infs and -infs. Then I want to replace the +inf entries with 0, the -inf entries with tau and other entries x with tau - x.

Since the covariance matrix is precomputed, instead of a theano Ops maybe it is easier to generate a index matrix outside of the pymc3 model, and assign 0, tau, tau-x to different index inside of the model using tt.inc_subtensor