How to use the model trained in Dopamine

Unity Obstacle Tower Challenge
#1

Hi,

I’ve worked through the Dopamine training guide and my model is currently training.

Thanks to the great resources provided, everything went smoothly even though I was working on a windows machine. I struggled a bit setting up port forwarding for tensorboard but this helped me do it through putty.

Once the model plateaus and I’m ready to use it in run.py, is there an implementation that was provided that will use the model to play the game or should we implement one ourselves?

I’ve taken a look at unity_lib.py provided in dopamine_otc.zip and that seems like the place I would start if I need to implement it myself, but I wanted to make sure I didn’t miss something before I started.

Thanks!

2 Likes
#2

Hi qualityLettuce,

We didn’t include an implementation for using Dopamine in run.py. But you’re right, it should be pretty easy to port over some of the Dopamine code for use in evaluation. In particular, I’d take a look at run_experiment.py, and specifically the methods _initialize_checkpointer_and_maybe_resume and _run_one_episode. You should be able to modify these to work in run.py - you should be able to use the unity_lib.py as-is.

1 Like
#3

Does anybody know of a suitable solution to run dopamine models in inference mode for the evaluation?
It looks like that dopamine does provide a solution out of a box, which is quite a shame for google.