RL Frameworks

Oct 30, 2019 1 min read

RL Frameworks and Reproducibility


stable-baselines	Yes	TF 1.14	Atari, MuJoCo	★★★★★	★★★★★
vel	Yes	PyTorch 1.0	Atari, MuJoCo	★★★	★★★★
coach	Yes	TF	Atari, MuJoCo, VizDoom, Fetch	★★★★★	★★★★★
rllib	Yes	TF, PyTorch	Atari, HalfCheetah	★★★★	★★★★
Garage	Yes	TF, PyTorch	Pendulum, Walker, HalfCheetah, Fetch, Hopper etc		★★★
anyrl-py	☹	TF	☹	★	★★
lagom	Yes	PyTorch	MuJoCo	★★★	★★★
lagom	Yes	PyTorch	MuJoCo	★★★	★★★
reaver	Yes	TF 1.1+	StarCraft II	★★★	★★★
TF-Agents	☹	TF	☹	★★	★★
p-christ/DRL	PyTorch	Four Rooms, Ant etc	★★	★★★
TorchRL	Basic	PyTorch	Cartpole	★★	★★
OpenSpiel	☹	TF 1.14	games	★★★	★★★★
bsuite	Yes	TF 1.13	custom	★★★	★★★★
dopamine	Yes	TF	Atari	★★★	★★★★
keras-rl	☹	TF	☹	★★★	★★★
trfl	☹	TF	☹	★★★	★★★
tensorforce	basic	TF 1.13	☹	★★★	★★★
SLM Lab	Atari, Roboschool	TF 1.13	☹	★★★	★★★★
Simple RL	☹	TF 1.13	☹	★★★	★★★★
Horizon/ReAgent	☹	☹	★★
Torchbeast	IMPALA/Atari	☹	★★★

A program trying to understand what it’s computing.