Is it possible to train a model in scala environment, serialize it with mleap and then serve on a different python server? I've read the docs, but it still isn't clear to me how compatible those things are
Asked
Active
Viewed 115 times
0
-
Usually, you do it the other way around. mleap helps you with serving models on non-python servers. can you please elaborate on your case? – orlevii Sep 30 '20 at 20:24
-
Our case is that nobody really knows scala enough to write model's backend in it. – Nurgling96 Oct 01 '20 at 11:05
-
@orlevii why do you want to serve model in non python environment tho? – Nurgling96 Oct 01 '20 at 11:06
-
Can I ask why is your model written in scala? Is it for using spark MLlib? – orlevii Oct 01 '20 at 11:32
-
I think it's much easier to use the same language for training and serving ML models, things are much more compatible this way. – orlevii Oct 01 '20 at 11:35
-
Hmm, I thought that mleap is not language-dependent. Isn't it? – Nurgling96 Oct 01 '20 at 12:04
-
It isn't language-dependent, mlleap allows you to serialize models trained with sklean/spark/tensorflow and load them with their SDK. Usually, data scientists go-to language is python, where their full stack of pandas, numpy and other utilities are available. But here you are trying to do the other way around, so I'm a little confused – orlevii Oct 01 '20 at 12:47