search

lucasmaystre / choix

Inference algorithms for models based on Luce's choice axiom
MIT License
157 stars 27 forks source link

Weights for the importance of each observation #22

Open amirbachar opened 2 years ago

amirbachar commented 2 years ago

A nice feature would be adding a weights vector to the data vector, in order to be able to assign a different "importance for each observation". Using this suggested feature, it would also be easy to implement a regularisation, by adding another observation for each pair (in the pairwise comparison models), and give it a small weight. Currently when the directed graph is acyclic, the ML is that the root will basically have an infinite strength, and regularisation fixes that.

lucasmaystre commented 2 years ago

Hi @amirbachar thank you for the suggestion.

Note that most functions have a parameter alpha that controls regularization. it is off (i.e., equal to 0.0) by default in most functions, but if you set it to some positive value the infinite strength problem should be resolved.

amirbachar commented 2 years ago

@lucasmaystre Thank you very much for your response!