Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 

71 行
1.5 KiB

behaviors:
SmallCubeSoldier:
trainer: ppo
batch_size: 2048
beta: 0.005
buffer_size: 20480
epsilon: 0.2
hidden_units: 512
lambd: 0.95
learning_rate: 0.0003
learning_rate_schedule: constant
max_steps: 1.0e8
memory_size: 128
normalize: false
num_epoch: 3
num_layers: 2
threaded: false
time_horizon: 1000
sequence_length: 64
summary_freq: 10000
use_recurrent: false
vis_encode_type: simple
reward_signals:
extrinsic:
strength: 1.0
gamma: 0.99
self_play:
window: 10
play_against_latest_model_ratio: 0.5
save_steps: 50000
swap_steps: 15000
team_change: 200000
curriculum:
measure: progress
thresholds: [0.025, 0.05]
min_lesson_length: 100
signal_smoothing: true
parameters:
bonus: [1.0, 0.5, 0.0]
LargeCubeSoldier:
trainer: ppo
batch_size: 2048
beta: 0.005
buffer_size: 20480
epsilon: 0.2
hidden_units: 512
lambd: 0.95
learning_rate: 0.0003
learning_rate_schedule: constant
max_steps: 1.0e8
memory_size: 128
normalize: false
num_epoch: 3
num_layers: 2
threaded: false
time_horizon: 1000
sequence_length: 64
summary_freq: 10000
use_recurrent: false
vis_encode_type: simple
reward_signals:
extrinsic:
strength: 1.0
gamma: 0.99
self_play:
window: 10
play_against_latest_model_ratio: 0.5
save_steps: 50000
swap_steps: 1700
team_change: 200000