Human-Level Control through Deep Reinforcement Learning

Research paper of Human-Level Control through Deep Reinforcement Learning.

This implementation contains:

1. Deep Q-network and Q-learning
2. Experience replay memory
   • to reduce the correlations between consecutive updates
3. Network for Q-learning targets are fixed for intervals
   • to reduce the correlations between target and predicted Q-values

Requirements

• Python 3.6
• gym
• tensorflow

how to use

Using config.json

    /*Config for CartPole-v0*/
    {   
    "env_type": "2D",                    // Env type: Dimensions of input game state 2D or 3D "Box2d has 2d env and Atari has 3d env but if you use atari ram has env type 2d"
    "game_name": "CartPole-v0",          // Game name
    "learning_rate": 0.0025,             // Learning Rate
    "decay_rate" : 1.001,                // Decay rate
    "num_of_game_play": 100000,          // Nmber of game play
    "learning_start_after_frames": 1000, // Number of random game play before learning
    "update_main_model": 1,              // update rate of your main model
    "update_target_model": 1,            // update rate of your target model
    "check_point" : "Check_point/Boxing-ram-v4_20200719184545_reward_80.0_frames_2816825.h5", // Check point path for resume or testing
    "render": true,                      // Gamplay desplay on screen or not 

    "training" : {                       // Training config
        "training": true,                // Want to train
        "save_model": false,             // Save Model??
        "resume": false,                 // Resume model form check_point path
        "batch_size": 32,                // Batch size
        "memory_max_size": 100000        // Memory for store frams
        
    },
    "testing":{                          // Testing config
        "testing": false
    }
}

• Using config play around different games

how to run

    python dqn.py

Results

• Result of training on cpu 2 hrs

• Result of training on cpu 4 hrs.

Acknowledgement

Beat Atari with Deep Reinforcement Learning! (Part 1: DQN)