search

anuragajay / decision-diffuser

279 stars 39 forks source link

Different results of halfcheetah #1

Open rainbow979 opened 1 year ago

rainbow979 commented 1 year ago

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. 1673933898018

WangYong-Design commented 1 year ago

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

Looomo commented 1 year ago

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 commented 1 year ago

same question

SpaceLearner commented 1 year ago

same question

xishuxishu commented 4 months ago

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. 1673933898018

hello，Have you solved this problem