CODEWITHSUNDEEP
rdeploymentmachine-learningdeep-learningh2o
JaKu
JaKu

Reputation: 1166

Is there a way to use saved model between different versions of H2O?

I have saved a trained model (deep net, but it is more general I think) in H2O. Now I want to load it by another instance of H2O and use it for scoring, but the problem is, that the version of H2O used for training (3.10.0.3) was different than the one I started the production cluster with (3.10.0.6). The error message is quite self-explanatory

ERROR MESSAGE:

Found version 3.10.0.3, but running version 3.10.0.6

Is there a way to migrate the saved model between versions? Or am I stuck with using the same version of H2O for training and scoring?

Upvotes: 5

Views: 1698

Answers (1)

Darren Cook
Darren Cook

Reputation: 28928

Yes, you are stuck using the same version for training and scoring. No migration route.

(You can export a model as a POJO, which can be bundled with the version of h2o-genmodel.jar that it needs. But that requires writing Java code to get the data in and results out, which is not ideal if you are using R code for data preparation.)

This has been discussed on the h2o-stream mailing list before, but I couldn't see a feature request ticket for it, so I just created one: https://0xdata.atlassian.net/browse/PUBDEV-3432

Upvotes: 5

Related Questions