search

Wikidata / soweego

Link Wikidata items to large catalogs
https://meta.wikimedia.org/wiki/Grants:Project/Hjfocs/soweego_2
GNU General Public License v3.0
97 stars 9 forks source link

Implement ensemble classifiers #372

Closed tupini07 closed 4 years ago

tupini07 commented 5 years ago

This PR implements the remaining ensemble classifiers (closes #305):

For the record, the results (evaluation and linking) of these classifiers have been posted in issue #359

It also introduces some refactoring of the classifier module (reordering of the classifier definitions)

Finally, the scripts used to extract the reports posted in #359 have been included in the scripts/linker folder.

tupini07 commented 5 years ago

@marfox thanks for the comprehensive review! I still haven't finished fixing all the issues you mentioned. I just wanted to let you know that I just included a new feature in this PR: now classifiers which use mlens under the hood (GatedClassifier and StackedClassigier) return the probability of a match instead of just whether it was a match or not. This has been done in 5318d5f