CosineRecommender isn't pickable

[marinkreso]

I created CosineRecommender model and fitted it to my dataset. When I try to pickle the fitted model with pickle.dumps(model) I get TypeError: no default __reduce__ due to non-trivial __cinit__ error in because of NearestNeighboursScorer.reduce_cython() method.

Pickling AlternatingLeastSquares model works great.

Is there a way to pickle CosineRecommender model or to use some other serialization method for serializing this object/model?

[mohjeih]

I have encountered recently the same issue as you. My quick solution which definitely is not the best one is as follows: assign model scorer to a temporary variable, assign 0 to the model scorer, pickle the fitted model, and finally assign the value of the temporary variable to the model scorer. Hopefully, this issue will be fixed in the future versions.

[benfred]

This PR adds support for pickling knn models: #229

[benfred]

fix is in v0.3.9

[marinkreso]

Hi, Thanks for great news. Best regards, Marin Krešo sub, 13. srp 2019. 18:52 Ben Frederickson <[email protected]> je napisao:

…

This PR adds support for pickling knn models: #229 <#229> — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#191?email_source=notifications&email_token=AEYHERND4WUFMW4TC4XIZU3P7IB4NA5CNFSM4G7JVSAKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODZ3VLDA#issuecomment-511137164>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEYHERI4KUW3VYBSDPR7IT3P7IB4NANCNFSM4G7JVSAA> .