It's actually not so gradiose.
Julia implementations of deep reinforcement learning algorithms. Uses the POMDPs.jl framework for representing (Partially Observable) Markov Decision Proccesses (which itself is a framework for describing sequential decision making problems).
Working-ish:
Currently working on:
To work on:
NOTE: the signature for solve doesn't exactly match POMDPs.solve
. It is `solve(::Solver, ::MDP, ::Policy, ::rng)
NOTE: try to define your vectors as Float32 if possible (which is what mxnet uses)
Documentation someday