The Article Recommendation Tool depends on scikit-learn, which has not been packaged for Python 3 in jessie. It is available in sid, though. To roll out ART in prod, python3-sklearn, python3-sklearn-lib, and python3-sklearn-pandas would need to be backported for jessie. This requires the assistance of someone in ops with packaging experience.
Description
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Resolved | leila | T112321 [Epic] Article recommendation productization/integration | |||
Declined | None | T133362 Backport python3-sklearn and python3-sklearn-lib from sid |
Event Timeline
I took a quick look at this, the source packages we'd need are: scikit-learn, joblib (dependency of the former) and sklearn-pandas. I think it'd make sense to upload them to jessie-backports once the packages and their deps are all in testing as per backports policy.
Why is packaging necessary? With ORES, we are using sklearn via a wheel rather than a Deb package. Could that work for deployments here too?
Per discussions in backlog grooming, there is no dependency on sklearn at the moment, however, for future experiments and developments, we may need this.
I also think deb packaging for this is going town a long, unrecoverable rabbit hole, and would recommend a wheels setup similar to what we have for ORES.
Per discussions in backlog grooming, there is no dependency on sklearn at the moment, however, for future experiments and developments, we may need this.
I'll remove the Blocked-on-Operations tag for now, then.