search

kiudee / cs-ranking

Context-sensitive ranking and choice in Python with PyTorch
https://cs-ranking.readthedocs.io
Apache License 2.0
66 stars 15 forks source link

Investigate GeneralizedNestedLogit threshold #158

Open timokau opened 4 years ago

timokau commented 4 years ago

Peculiar constant - this should rather be something passed into __init__ as parameter. But this is something for a separate issue.

_Originally posted by @kiudee in https://github.com/kiudee/cs-ranking/pull/157#discussion_r488695680_

prithagupta commented 4 years ago

We had to set this threshold, after trial and error. Because the generalized nested logit model has so many exponential terms that need to be evaluated for getting the final choice probabilities, that it used to run in some memory issues and for larger datasets. Since we cannot remove the features, we had to subsample the instances. This threshold is library dependent. Maybe we need to talk about using a different library for it. Or we can recheck if the model works with larger datasets in new versions of pymc3.

timokau commented 4 years ago

Thanks for the context @prithagupta :)