search

google-deepmind / bsuite

bsuite is a collection of carefully-designed experiments that investigate core capabilities of a reinforcement learning (RL) agent
Apache License 2.0
1.51k stars 181 forks source link

How is the 'generalization' score computed? #19

Closed VincentLiu3 closed 4 years ago

VincentLiu3 commented 4 years ago

In the notebook (bit.ly/bsuite-agents), I only found the description of 6 scores (basis, noise, scale, exploration, memory, and credit assignment). I wonder how is the generalization score computed? Thank you!

iosband commented 4 years ago

Each experiment is associated with a different "tag" for the purposes of the spokes. You can see the TAGS variable in the analysis.py for each experiment.

Then, the score for each spoke is given by the average by each tag: https://github.com/deepmind/bsuite/blob/master/bsuite/experiments/summary_analysis.py#L163