Skip to content

Latest commit

 

History

History
116 lines (114 loc) · 2.99 KB

Setup-Configuration-Files.md

File metadata and controls

116 lines (114 loc) · 2.99 KB
Raw

Setup Configuration Files

Training Config

Located in config/trainer_config.yaml
This is the configuration about Reinforcement Learning PPO trainer

AutoBenchBrain:
    batch_size: 1024
    beta: 1.0e-1
    buffer_size: 1024
    epsilon: 0.2
    gamma: 0.99
    hidden_units: 128
    lambd: 0.95
    learning_rate: 3.0e-4
    max_steps: 1.0e7
    memory_size: 256
    normalize: true
    num_epoch: 5
    num_layers: 2
    time_horizon: 512
    sequence_length: 64
    summary_freq: 3000   
    use_recurrent: true

Environment Config

Located in config/curricula/autobench/AutoBenchBrain.json This is about the configuration of Unity environment

{
  "measure": "progress",        #Ignore
  "thresholds": [],             #Ignore
  "min_lesson_length": 0,       #Ignore
  "signal_smoothing": false,    #Ignore
  "parameters": {
    "camera1_type": [0],
    "camera2_type": [3],
    "camera3_type": [0],
    "camera1_res_x": [0],
    "camera2_res_x": [50],
    "camera3_res_x": [0],
    "camera1_res_y": [0],
    "camera2_res_y": [50],
    "camera3_res_y": [0],
    "weather_id": [1],
    "time_id": [9],
    "road_width": [7],
    "forward": [true],
    "detail": [false],
    "goal_reward": [500],
    "time_penalty": [-1],
    "collision_penalty": [-300],
    "position_reward": [300],
    "velocity_reward": [1]
  }
}

Only need to focus on parameters section

Python Script

Located in learn_rl.py, learn_ml.py, learn_gym.py
The following uses learn_rl.py as an example

env_path = 'AutoBenchExecutable/AutoBenchExecutable' #Default executable path
run_id = '1'
load_model = False
train_model = True
save_freq = 10000
keep_checkpoints = 1000
worker_id = 0
run_seed = 0
curriculum_folder = 'config/curricula/autobench/'
curriculum_file = 'config/curricula/autobench/AutoBenchBrain.json'
lesson = 0
fast_simulation = True
no_graphics = False
trainer_config_path = 'config/trainer_config.yaml'
benchmark = False
benchmark_episode = 100
benchmark_verbose = True

Env_path

Path of the Unity executable

Run_id

Identifier for each run, suitable for fine-tunning parameters

Load_model

Whether load the tensorflow model

Train_model

Whether train the tensorflow model

Save_freq

Frequency of the tensorflow model saved

Keep_checkpoints

Maximum checkpoint allow for saving

Worker_id

Ignore and set to 0

Run_seed

Random seed of the Unity executable

Curriculum_folder

Folder of environment config file

Curriculum_file

Location of environment config file

Lesson

Ignore and set to 0

Fast_simulation

If set to True, small window, 100X time scale, 10 agents
If set to False, large window, 1X time scale, 1 agent and WASD-controled Observe Camera

No_graphic

Whether not showing the windows of Unity environment

Trainer_config_path

Location of trainer config file

Benchmark

Whether benchmark the current model

Benchmark_episode

Number of episode needed for benchmarking

Benchmark_verbose

Whether or not print out episode information if episode ends