This is a LXH-unfriendly, but YGY-friendly and GCX-CrazyHappy project. Unstable Baselines is designed to provide a quick-start guide for Reinforcement Learning beginners and a codebase for agile algorithm development. In light of this, only the basic version of each algorithm is implemented here, without tedious training skills and code-level optimizations. UB is currently maintained by researchers from lamda-rl, and a pypi source will be available once it is ready for publishing.

• Deep Q Learning (DQN)
• Deep Deterministic Policy Gradient (DDPG)
• Soft Actor Critic (SAC)
• Twin Delayed Deep Deterministic policy gradient algorithm (TD3)

• Randomized Ensembled Double Q-Learning (REDQ)
• Proximal Policy Optimization (PPO)
• Model-based Policy Optimization (MBPO)
• Efficient Off-policy Meta-learning via Probabilistic Context Variables (PEARL)

git clone --recurse-submodules https://github.com/x35f/unstable_baselines.git
cd unstable_baselines
conda env create -f env.yaml 
conda activate rl_base
pip3 install -e .

python3 /path/to/algorithm/main.py /path/to/algorithm/configs/some-config.json args(optional)

#install metaworld for meta_rl benchmark
cd envs/metaworld
pip install -e .
#install atari
pip install gym[all]