Chris MacLellan
Chris MacLellan
You know, I can't remember. I think it was a file that I got from one of my social media mining classes. However, I totally agree-- these words shouldn't be...
Note, this is a bug because the current configuration prevents the viz from working locally at the moment.
might try monkey patching to add hash caching to tuples just to see if there is any value
An alternative idea is to add a new kind of feature that supports something like numpy arrays directly. Now that I'm thinking about it this might be the best way...
at least pruning at recall.
The general idea here would be to treat the "not" relation in a special way. Effectively distributing its probability mass over all values not equal to the value specified.
There may be situations where this is actually desirable. Mainly, going all the way to a leaf might be useful for implementing Labyrinth or something.
Also, maybe we should go all the way to a leaf, then when making predictions decide which concept in the entire path to use for prediction.
So it looks like we just always go to the best leaf, all the way down the tree. See: https://github.com/cmaclell/concept_formation/blob/6610391ddb2844734d4b0b8baff92bc6c67e2777/concept_formation/cobweb.py#L203-L216
I think this makes sense for categorization, in that we can't create anything new, so we just find the best concept to put it in.. however, this never stops at...