Skip to content
  • No due date Last updated over 7 years ago

    formally defines action spaces and state spaces adds a universe en…

    • formally defines action spaces and state spaces
    • adds a universe environment
    • adds a roboschool environment
    • removes the requirement of manually spinning up a webserver to interact with gym environments.

  • No due date Last updated over 7 years ago

    (envs) Possibly split reinforce into third-party dependencies (ie: -…

    • (envs) Possibly split reinforce into third-party dependencies (ie: -envs-gym, -envs-atari)
    • (envs) Introduce some kind of History monad to cleanly seperate debug logging from agent or environment reporting
    • (algos) write out more baseline algorithms in reinforce-algorithms (in progress)
    • (algos) modify project structure to so that reinforce and reinforce-algorithms are siblings

    No hackage release with this. This would basically prototype a reinforce-algorithms project.

  • No due date Last updated over 7 years ago

    (envs) Have some kind of reporting system (live reporting?) or ser…

    • (envs) Have some kind of reporting system (live reporting?) or server in place for model inspection - similar to tensorboard.
    • (algos) port all v0.2.0 python prototypes into backprop in reinforce-deep-algorithms
    • (algos) port all v0.2.0 python prototypes into tensorflow-haskell in reinforce-deep-algorithms

    Optionally look into grenade. Figure out if we want to put this in reinforce-algorithms, reinforce-zoo, or something else.