mirror of
https://github.com/wassname/ray.git
synced 2026-08-10 12:30:14 +08:00
* Bug fix: DQN goes into negative epsilon values after reaching exploration percentage. * Add `epsilon_initial_eps` to SAC to pass test_nested_spaces.py. * Add `exploration_initial_eps` to QMIX default config.
Implementation of Soft Actor-Critic (https://arxiv.org/abs/1812.05905.pdf).