search

RPegoud / jym

JAX implementation of RL algorithms and vectorized environments
MIT License
32 stars 2 forks source link

Add epsilon annealing #18

Closed RPegoud closed 11 months ago

RPegoud commented 11 months ago

Currently, the training of DQN on Cartpole is quite unstable, probably due to a constant epsilon value. To test performances, add epsilon annealing.

image image