search

amharrison / jactr

-old - jACT-R Bundles
http://jact-r.org/
7 stars 5 forks source link

Add support for additional filtering of productions to be rewarded or compiled in proc learning module #21

Closed amharrison closed 8 years ago

amharrison commented 8 years ago

Modeler might need to exclude individual productions from utility of compilation. Current filtering based on buffers is not always sufficient.

It should be adequate to provide a filtering chain that users can manipulate for their own purposes.

amharrison commented 8 years ago

Or, we can just tweak the behavior of the 'Reward' parameter. Currently it is only numeric reward value, or NaN, if none. We can add a keyword value 'skip' (case insensitive), that will do just that, but allow the reward propagation to continue.

This solution requires no API access, which makes it preferable in my eyes. Plus, it's really clear what the intent is. Perhaps we should alias 'NaN' with 'default'? And maybe provide 'stop' for terminating the reward chain?

amharrison commented 8 years ago

Need to resolve https://github.com/amharrison/jactr/issues/22 first.

amharrison commented 8 years ago

Implemented with #22. Available in the latest dev release