pytorch-drl-algs

Model Free Deep Reinforcement Learning Algorithms

Algorithms:

• Rainbow [code] [tests]
  • DQN and
  • DDQN or
  • Dueling DQN or
  • Noisy Networks or
  • Prioritized Experience Replay or
  • N Step DQN or
  • Categorical DQN / Quantile Regression
• DDPG [code] [tests]
• TD3 [code] [tests]
• SAC [code] [tests]
• ACER [code] [tests]
• A2C [code] [tests]
• TRPO [code] [tests]
• ACKTR [code] [tests]
• PPO [code] [tests]
• GAIL [code] [tests]
• HER [code] [tests]
• MADDPG [code] [tests]

Note: Not tested thoroughly. Not to be used for anything serious

---

---

Packages

• pytorch == 1.5
• gym == 0.15.6
• multiagent particle environment (for MADDPG)
• numpy == 1.18

---

---

Acknowledgements/References

• ACKTR: KFAC implementation of @ikostrikov
• TRPO: For line search this is used as reference. @ikostrikov
• Noisy Linear Layers @higgsfield
• Parallel Environment. @dolhana
• For PER: segment tree @hill-a
• For policy_heads: distribution modules of @ikostrikov is used.