You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
So there is an inherent contradiction in this in that looking at the past examples is not really incremental anymore. We could have some kind of an n-back or mini-batch how learning or something to make it more incremental.
This being said, I think it should be pretty straightforward-- probably more so with Danny's refactor. If we have an interface for the how learner, it should be able to accept more than one example and it should try to find a solution that covers them all. Given that a solution that covers them all has to be a solution to each of them, we can just do the regular search for one of the examples (the most recent?) and just reject solutions that do not cover all the past positive examples as well.
Activity
eharpste commentedon Apr 8, 2019
Could apply this within all examples provided to the same skill label.
cmaclell commentedon Apr 15, 2019
So there is an inherent contradiction in this in that looking at the past examples is not really incremental anymore. We could have some kind of an n-back or mini-batch how learning or something to make it more incremental.
This being said, I think it should be pretty straightforward-- probably more so with Danny's refactor. If we have an interface for the how learner, it should be able to accept more than one example and it should try to find a solution that covers them all. Given that a solution that covers them all has to be a solution to each of them, we can just do the regular search for one of the examples (the most recent?) and just reject solutions that do not cover all the past positive examples as well.
eharpste commentedon Jun 3, 2019
This is related to #32. Is it also related to #26?