search

NTMC-Community / MatchZoo

Facilitating the design, comparison and sharing of deep text matching models.
Apache License 2.0
3.82k stars 898 forks source link

When to expect tutorial for anmm model and listwise loss? #782

Open ctrado18 opened 4 years ago

ctrado18 commented 4 years ago

Hi,

I just want to know If you can expect a tutorial in wikiqa for the anmm model like for the others to? Otherwise, It is a bit too hacky to implement this by myself.

Also, is it right now possible to use a listwise loss like the pairwise loss with any model? I want to train some of the model in the tutorial like arci for a listwise loss. Is this possible right now? I want to mark relevant, and also partial relevant as also non relevant documents with this loss.

matthew-z commented 4 years ago

IIRC, Rank cross entropy loss is a kind of listwise loss

uduse commented 4 years ago

I don't think it will be added anytime soon.

yangliuy commented 4 years ago

Hi @ctrado18, thank you for your message!

You can refer to the performance of aNMM model on WikiQA data here https://github.com/NTMC-Community/MatchZoo/blob/1.0/README.md . The implementation of aNMM based on Keras is here https://github.com/NTMC-Community/MatchZoo/blob/1.0/matchzoo/models/anmm.py. Hope this helps.