Closed keryums closed 3 years ago
Hi @keryums
maybe this excerpt from the notes in the documentation is helpful:
If no feature matrices are provided to the lightfm.LightFM.fit() or lightfm.LightFM.predict() methods, they are implicitly assumed to be identity matrices: that is, each user and item are characterised by one feature that is unique to that user (or item). In this case, LightFM reduces to a traditional collaborative filtering matrix factorization method. https://making.lyst.com/lightfm/docs/lightfm.html
If you pass in features to the fit calls and you have used the built-in dataset construction, than adding one "identity feature" per user and item is also the default. See: https://making.lyst.com/lightfm/docs/lightfm.data.html#lightfm.data.Dataset
I hope this helps!
If this doesn't clear things up, maybe you could provide a concrete example, e.g. shape of your matrices, way of constructing them.
Thank you that makes sense!
Ker-Yu 415-312-2148
On Wed, Dec 30, 2020 at 4:22 PM Simon Weiß notifications@github.com wrote:
Hi @keryums https://github.com/keryums
maybe this excerpt from the notes in the documentation is helpful:
If no feature matrices are provided to the lightfm.LightFM.fit() or lightfm.LightFM.predict() methods, they are implicitly assumed to be identity matrices: that is, each user and item are characterised by one feature that is unique to that user (or item). In this case, LightFM reduces to a traditional collaborative filtering matrix factorization method. https://making.lyst.com/lightfm/docs/lightfm.html
If you pass in features to the fit calls and you have used the built-in dataset construction, than adding one "identity feature" per user and item is also the default. See: https://making.lyst.com/lightfm/docs/lightfm.data.html#lightfm.data.Dataset
If this doesn't clear things up, maybe you could provide a concrete example, e.g. shape of your matrices, way of constructing them.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/lyst/lightfm/issues/577#issuecomment-752679445, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFJBSJPSJCBE7XSWRPSVRUDSXNHT5ANCNFSM4VOG3AKA .
Perfect, then I'll close.
Hello,
How does this package deal with the case of passing in item features for fewer items than is in the user-item utility matrix? We've noticed that the model will train and we can extract item representations from the trained model even without explicitly imputing for these missing item features.
Thanks