search

nottombrown / rl-teacher

Code for Deep RL from Human Preferences [Christiano et al]. Plus a webapp for collecting human feedback
MIT License
559 stars 93 forks source link

Video clips not showing when providing labels by human #26

Closed Guzzii closed 7 years ago

Guzzii commented 7 years ago

Followed through the instructions. Everything seems work well. Only one issue is that I cannot get the video clips to show up on the label page. I am on macOS Sierra. Any thoughts? Thanks

nottombrown commented 7 years ago

Hey. Is the data correctly being stored in GCS? I would recommend running

gsutil ls <your bucket>

And make sure that the files are actually uploaded. If they are not then you may need to fix some bucket permissions.

Guzzii commented 7 years ago

Hi @nottombrown . Thank you very much for your promptly reply. You're right. The files were not uploaded to GCS. This is what I got,

$python rl_teacher/teach.py -p human --pretrain_labels 175 -e Reacher-v1 -n human-175
Using TensorFlow backend.
No label limit given. We will request one label every few seconds
Starting random rollouts to generate pretraining segments. No learning will take place...
Collected 10/350 segments
Collected 20/350 segments
Collected 30/350 segments
Collected 40/350 segments
Collected 50/350 segments
Collected 60/350 segments
Collected 70/350 segments
Collected 80/350 segments
Collected 90/350 segments
Collected 100/350 segments
Collected 110/350 segments
Collected 120/350 segments
Collected 130/350 segments
Collected 140/350 segments
Collected 150/350 segments
Collected 160/350 segments
Collected 170/350 segments
Collected 180/350 segments
Collected 190/350 segments
Collected 200/350 segments
Collected 210/350 segments
Collected 220/350 segments
Collected 230/350 segments
Collected 240/350 segments
Collected 250/350 segments
Collected 260/350 segments
Collected 270/350 segments
Collected 280/350 segments
Collected 290/350 segments
Collected 300/350 segments
Collected 310/350 segments
Collected 320/350 segments
Collected 330/350 segments
Collected 340/350 segments
Collected 350/350 segments
Successfully collected 350 segments
0/175 comparisons labeled. Please add labels w/ the human-feedback-api. Sleeping...
0/175 comparisons labeled. Please add labels w/ the human-feedback-api. Sleeping...
0/175 comparisons labeled. Please add labels w/ the human-feedback-api. Sleeping...

For some reason, it bypassed the uploading process?? Do I need to specifically run video.py to upload?

nottombrown commented 7 years ago

Hmm, I can't replicate this. When I run your command, I get the same output that you described, and then I get the following logs describing successful uploads.

Copying media to gs://testo/5bd806a1-5344-4a07-bb37-0e37065f4f85-right.mp4 in a background process

Can you manually upload files using the command in video.py?

nottombrown commented 7 years ago

Closing this as stale