Model serving

[ghost]

Is there a way to serve the models created with TransmogrifAI?

I found PredictionIO useful but it seems that that project was put in the Apache attic.

[tovbinm]

We support the serving models using Spark and MLeap. With Spark, simply run model.score(..). For MLeap, have a look here - https://github.com/salesforce/TransmogrifAI/tree/master/local

[ghost]

thank you @tovbinm
I see, so it is possible to export the model in MLeap format and then use MLeap serving to serve it in production.

[leahmcguire]

Only stages wrapping spack will be saved as mleap (we have not written converters for stages written in transmogrifai) so it will need to be loaded as a local model in transmogrifai https://github.com/salesforce/TransmogrifAI/tree/master/local

[leahmcguire]

Though we don't have any utils for packaging it as an endpoint if that is what you are asking.

[ghost]

Yes, the question was a bit ambiguous @leahmcguire
I was looking for some tools/utils to serve the serialized model in production.
So far I used to use PredictionIO.