Super-confident predictions

[marfox]

Different classifiers may capture different relations in the data. We can join what each classifier learns by creating ensembles of them.

• [x] Use sklearn's VotingClassifier instead of our implementation - #366
• [x] Implement a Gating ensemble - #356
• [x] Implement a Stacking ensemble - #367

Results will then be presented in a separate task

• [x] Analyze performance of ensemble methods - #359

[tupini07]

It seems that ensembles are commonly used in RL pipelines to predict if 2 entities match or not. Ensembles are usually composed of diverse models (eg, SVM + Decision Trees). A couple of examples:

In [1] a final ensemble is used to self learn on a partially labelled dataset. In [2] they use multiple classifiers to predict a match, the final classifier is selected depending on its score and how interpretable the model is.

1. A novel ensemble learning approach to unsupervised record linkage (2017)
2. Magellan: Toward Building Entity Matching Management Systems (2016)