feat: 优化DQN训练配置并支持Dueling网络评估

Browse Source

- 将学习率调整为5e-5，批次大小增加到64，经验回放缓冲区扩大到500,000
- 启用优先经验回放，调整目标网络更新频率为1000步
- 评估时使用Dueling网络架构，训练时评估模式的ε设为0
- 更新评估结果以反映配置改进后的性能变化

...

This commit is contained in:

刘航宇

2026-05-02 11:36:12 +08:00

parent faf0d5ea42

commit 79ffb90823

11 changed files with 58 additions and 46 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

强化学习个人项目报告（Atari 游戏方向）/models/dqn_step_2000000.pt

BIN

Binary file not shown.