Different results of halfcheetah

[rainbow979]

Hello,

I tried the same config with the repo and got the same good performance with the paper. However, when I tried the env halfcheetah and the testing score is much lower than the results from the paper. I only changed the returns_scale from 400 to 800 since halfcheetah had higher discounted returns. The training loss is shown in the figure. The blue line is halfcheetah and the purple line is hopper. I am wondering if there are other hyperparemeters needs to be changed.

[WangYong-Design]

Have you kept track of how long it takes to run an experiment on GPU?

[Looomo]

Hi, I was trying to reproduce the results of DD, but I couldnt make it, results on each dataset differ significantly from the results presented in the paper in Tab1. （I have tried only on 4 datasets, walker2d-medium-replay-v2, walker2d-medium-v2,halfcheetah-medium-replay-v2, and halfcheetah-medium-v2）. So have you change anything in this repo or just directly run the downloaded code? Thanks!
Also, did you find the codes of length-K history condition in this repo?

[RenMing-Huang]

same question

[SpaceLearner]

same question

[xishuxishu]

Hello,

I tried the same config with the repo and got the same good performance with the paper. However, when I tried the env halfcheetah and the testing score is much lower than the results from the paper. I only changed the returns_scale from 400 to 800 since halfcheetah had higher discounted returns. The training loss is shown in the figure. The blue line is halfcheetah and the purple line is hopper. I am wondering if there are other hyperparemeters needs to be changed.

hello，Have you solved this problem

[wangerlie]

Hello,

I tried the same config with the repo and got the same good performance with the paper. However, when I tried the env halfcheetah and the testing score is much lower than the results from the paper. I only changed the returns_scale from 400 to 800 since halfcheetah had higher discounted returns. The training loss is shown in the figure. The blue line is halfcheetah and the purple line is hopper. I am wondering if there are other hyperparemeters needs to be changed.

hello @rainbow979
I also tried to reproduce the results, I run the code directly without ANY change in the code, I just couldn't get the same result in 'hopper-medium-expert-v2'. I am wondering how to reproduce the result in 'hopper-medium-expert-v2', have you ever change any hyperparameters? Thanks a lot .

[wangerlie]

I am also wondering do you have the same warning as following, I think it's not reasonable, there may have some problem within the normalization code?