search

dfridovi / rl

A homebrewed C++ library for reinforcement learning.
https://dfridovi.github.io/rl
Other
0 stars 0 forks source link

Fixed Q targets #7

Closed dfridovi closed 7 years ago

dfridovi commented 7 years ago

Implement fixed Q targets idea, i.e. train towards a static Q function for a bunch of iterations, then copy new weights over and repeat. Note that this will involve changing two things:

  1. Change the ContinuousQLearning<StateType, ActionType> class to use two Q functions, and
  2. Make sure that each child class of ContinuousActionValueFunctor<StateType, ActionType> has a working copy contructor (hopefully the default one is good enough...)
dfridovi commented 7 years ago

(2) is done.

dfridovi commented 7 years ago

(1) is done too. Seems to work!