search

krezelj / academia

Repository for an engineering thesis at Warsaw University of Technology that touches on the topic of curriculum learning.
MIT License
7 stars 0 forks source link

Experiment 7 #204

Closed krezelj closed 8 months ago

krezelj commented 9 months ago

This is pretty much the same script I've used for other experiments but with some key changes so pleace review carefully. I've also added a run_offset option which can be used to easily coordinate experiment runs across several machines. Ideally the run offset should be a multiple of 5 as I've set the maximum number of runs to 5 per machine (5 per curriculum and 5 per no curriculum). I think it's plenty enough. Since it doesn't really matter I suggest that when running the experiment I set the run_offset option to 0, Szymon sets it to 5 and Maciej sets it to 10. If you have any questions feel free to ask :)

Example command to run the script python run.py -r 1 -v 2 -o <your_offset_value> -a ppo

krezelj commented 9 months ago

One more thing, for now I suggest we only run using PPOAgent. If we have time left we can use DQNAgent as well.

krezelj commented 9 months ago

A command I'm using to run 1 curriculum and 1 no curriculum overnight python run.py -r 1 -v 2 -o 0 -a ppo -dn; python run.py -r 1 -v 2 -o 0 -a ppo -dc;

maciejors commented 8 months ago

@krezelj is this ready to merge or is there anything else to add here?

krezelj commented 8 months ago

Sorry it slipped my mind, merged!