default instructions for monobeast in Pong

[sunchipsster1]

Hello! I was wondering how the plots (attached) from this repo were produced? Was it done on Polybeast or Monobeast?

The reason I ask is that when I followed the default instructions:

python -m torchbeast.monobeast --env PongNoFrameskip-v4

--> Pong (which should ordinarily be an easy game) doesn't learn anything and reward remains at -20 even after millions of timesteps. Do you potentially have insight about what is going on?

Thank you so much in advance for your kind help!

[heiner]

Hey!

These plots were generated with PolyBeast. We never did the gridsweep necessary to find good hyperparameters for monobeast.

Polybeast is a bit of a ... beast .. to install. These days, I'd suggest taking a look at moolib.