marlbenchmark / on-policy

This is the official implementation of Multi-Agent PPO (MAPPO).
https://sites.google.com/view/mappo
MIT License
1.27k stars 292 forks source link

模型性能问题 #91

Closed bitbjt closed 11 months ago

bitbjt commented 11 months ago

请问如何评估训练出的模型性能好坏呢?输出的actor和critic模型是要通过eval来检验吗?

zoeyuchao commented 11 months ago

是的,通常来说,提前设定好评价指标,然后eval对应模型的结果。

发自我的iPhone

在 2023年10月29日,16:25,Shang Wenlong @.***> 写道:

 请问如何评估训练出的模型性能好坏呢?输出的actor和critic模型是要通过eval来检验吗?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.

bitbjt commented 11 months ago

是的,通常来说,提前设定好评价指标,然后eval对应模型的结果。 发自我的iPhone 在 2023年10月29日,16:25,Shang Wenlong @.***> 写道:  请问如何评估训练出的模型性能好坏呢?输出的actor和critic模型是要通过eval来检验吗? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.

好的谢谢,这里的代码需要自己补充吗?

zoeyuchao commented 11 months ago

现在repo里有对应的eval脚本。

On Mon, Oct 30, 2023 at 1:58 PM Shang Wenlong @.***> wrote:

是的,通常来说,提前设定好评价指标,然后eval对应模型的结果。 发自我的iPhone … <#m2305915839063829792> 在 2023年10月29日,16:25,Shang Wenlong @.***> 写道:  请问如何评估训练出的模型性能好坏呢?输出的actor和critic模型是要通过eval来检验吗? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.

好的谢谢,这里的代码需要自己补充吗?

— Reply to this email directly, view it on GitHub https://github.com/marlbenchmark/on-policy/issues/91#issuecomment-1784531766, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACF4B5UWQC5FWS4RQPKPD23YB46YHAVCNFSM6AAAAAA6UVWN46VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTOOBUGUZTCNZWGY . You are receiving this because you commented.Message ID: @.***>

--

Regards,

Chao Yu Ph.D TEL: +86-1565-2601-306 E-mail: @.** Homepage: https://nicsefc.ee.tsinghua.edu.cn/people/ChaoYu https://nicsefc.ee.tsinghua.edu.cn/people/ChaoYu* Address: Shuangqing 5302, Department of Electronic Engineering, Tsinghua University, Beijing 100084, P.R.China