JoeyAndres / rl

Reinforcement Learning library
2 stars 0 forks source link

Gradient Descent step size is divided by number of tilings. #33

Closed JoeyAndres closed 7 years ago

JoeyAndres commented 7 years ago

Currently GradientDescent's stepSize is stepSize/numTiling which is I believe I copied from Richard Sutton's implementation some time ago. I think this design is random. I believe the developer can set this to his/her specification. Having this blocks #32 since SDM don't have a concept of tiling.