I'd like to be able to reinflate the model from json, or deserialize it, so that it can be used for prediction estimation. For now I'm dealing exclusively with sklearn. I've been successful with pickle serializing + deserializing models, so they can be used for s SVC related predictions. These I think would be the easiest of a custom json encoder would to be created? Has anyone created a custom json encoder + decoder for SVC like SVM or SVR. And to continue, is it possible to create a custom encoder for ensemble methods like bagging, since it's a little more randomized internally?
Thank you, Jeff Levesque https://github.com/jeff1evesque > On Feb 5, 2018, at 11:49 PM, Joel Nothman <joel.noth...@gmail.com> wrote: > > I think you need to describe what use cases you intend once you've encoded > the thing. > > JSON's pretty generic. You can convert any pickle into JSON, but it'll still > have the security and versioning issues of a pickle. You can convert to PMML > and convert the XML to JSON, but it'd still be limited by what PMML can > represent, and not be deserializable into a sklearn estimator. > > Are you archiving the model? Are you hoping to predict from the model? Are > you hoping to continue training the model? Does it need to be deserializable > to a sklearn estimator? To a generic estimator not in the central sklearn > library? > > _______________________________________________ > scikit-learn mailing list > scikit-learn@python.org > https://mail.python.org/mailman/listinfo/scikit-learn
_______________________________________________ scikit-learn mailing list scikit-learn@python.org https://mail.python.org/mailman/listinfo/scikit-learn
