melli0505/deep-rl

Deep Reinforcement Learning with Custom Environment

Deep Reinforcement Learning with Custom Environment

This repository includes various Deep Reinforcement learning model training with a custom environment.

• I created a custom model for my case using the gym library and modified some model structures and training sequences.
• I haven't implemented complete models. Each model structure and wrapper have their own repositories or references, which I have mentioned in each model's README.

Custom Environment

• The Custom Environment simulates a Virtual Temperature Control System. Each action represents a change in temperature after performing an virtual control action (not implemented, you can insert your own real-world simulation environment).
• The target temperature in this system ranges between 20-25. I've set different rewards based on differences between the state (temperature) and the target temperature. You can adjust the reward system to match your own system.
• I haven't implemented the render and close functions in the environment object. If needed, you can refer to examples in the original gym repository (https://github.com/openai/gym/tree/master/gym/envs).

DQN with keras-RL Agent

Apply keras-rl DQN Agent(https://github.com/keras-rl/keras-rl/blob/master/rl/agents/dqn.py) to custom virtual Temperature control environment.

• Reference - https://www.section.io/engineering-education/building-a-reinforcement-learning-environment-using-openai-gym/

Theory(in korean)

• Markov Decision Process & Q-Learning(https://dnai-deny.tistory.com/80)
• DQN (https://dnai-deny.tistory.com/81)

DDPG with tensorflow

Deformed DDPG Agent and Actor-Critic models that implemented by @philtabor. Original Repository and Code is in https://github.com/philtabor/Youtube-Code-Repository.

Dueling DQN with tensorflow

Deformed Dueling DQN models that implemented by @philtabor. Original Repository and Code is in https://github.com/philtabor/Youtube-Code-Repository.

I set the effect of reward on training step 3 times larger than normal one. You can check it out in dueling_dqn.py, line 65.

For more information, see DuelDQN_custom/README.md

Dueling Double DQN with tensorflow

Deformed Dueling DQN models that implemented by @philtabor. Original Repository and Code is in https://github.com/philtabor/Youtube-Code-Repository.

I set the effect of reward on training step 10 times larger than normal one. You can check it out in dddqn.py, line 59 - 61.

For more information, see DDDQN_custom/README.md