search

opencog / rocca

Rational OpenCog Controlled Agent (ROCCA). Use OpenCog to control a rational agent in OpenAI Gym and Malmo environments.
GNU Affero General Public License v3.0
32 stars 18 forks source link

Add user parameter to weight the probability estimate of success #66

Closed ngeiswei closed 2 years ago

ngeiswei commented 2 years ago

Add a user parameter i to use

p * w^i

instead of p as probability of success estimate to select the best action, where w is the posterior second order probability factor.

Not sure how to conceptually justify this but it should offer some more room to control exploration vs exploitation, which is proving difficult.

The PR also contains various logging improvements.

ngeiswei commented 2 years ago

Weird, ci complains about black formatting, yet black does not complain on my system (my version of black is 21.12b0). Any idea what could be going wrong, @ntoxeg?

ngeiswei commented 2 years ago

I'm merging now to be able to move on.

ntoxeg commented 2 years ago

@ngeiswei you should probably update black, especially that it’s out of beta now.

ngeiswei commented 2 years ago

Thanks @ntoxeg!