SAC HER example results not matching

[Shade5]

I cloned the repo, setup the environment and ran (made no changes)

python her_sac_gym_fetch_reach.py

The results don't seem to match with this. Did something break in the latest commit?

However, when I try the td3, it works fine

python her_td3_multiworld_sawyer_reach.py

[vitchyr]

Can you run multiple seeds? I included 5 separate runs since it does tend to fail sometimes.

[Shade5]

Sorry, but I can't seem to find the right place to set the seeds. Would you link me to a line number?

[vitchyr]

It should be set randomly so you can just run the script again.

On Mon, Apr 22, 2019, 2:49 PM George Joseph notifications@github.com wrote:

Sorry, but I can't seem to find the right place to set the seeds. Would you link me to a line number?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/vitchyr/rlkit/issues/49#issuecomment-485564749, or mute the thread https://github.com/notifications/unsubscribe-auth/AAJ4VZOCBU2VVJOXJZJ5JZDPRYXIJANCNFSM4HHQCCRA .

[Shade5]

Yes, some of them work now, thanks.

[Shade5]

On a side note, I was trying FetchPush-v1. The only changes I need to make is in gym.make(... ) right?

If so, I've tried multiple seeds with about 500 iterations, with no progress. Am I missing something? What's the expected time?

[vitchyr]

Based on this paper, it seems like you need one millions steps, so probably (1000 iterations).

But they also use DDPG with an epsilon greedy exploration policy, so that might matter.

[Shade5]

For other users to confirm their results, here's FetchPush-v1 for 3 seeds.

Here's the results for 3 seeds on FetchPickAndPlace-v1