Spark 2.2. makes DL possible in Production as per Databricks

Seems interesting to me

That does look pretty good. Have you tried it?

No just saw this today. There have been threads in the forums where people were discussing how to use DL in production. So I thought this would be relevant. Maybe I should have just posted it in those threads.

Databricks has always seemed to me like an Enterprise thing - you have to buy in all of it. But that may be a perception thing.I have not thoroughly checked their offering. But this made it seem like this is possible in Apache Spark itself. I was thinking if someone had more information.

BTW for people who are interested Google Cloud allows offers hosting tensorflow models for predictions. They have something called ML Engine to train models. If you have a tensorflow models then you can also host them for predictions. I think they also support Keras.

Also there has been some news that AWS is trying to better support tensorflow. I think they are trying to compete with Google so that people do not only consider to go with Google for tensorflow. Given the recent tie ups between AWS and Microsoft to create Gluon I think we may see some relevant announcements in the coming AWS re:Invent in the next few weeks and maybe the Microsoft’s event whenever that is (I don’t keep tabs on Microsoft much).