What are the options for deploying models in production with R?

Question

According to my notice, there don't seem to be much options for deploying predictive models in production. 

I am in knowledge that PMML can be used to export models as an XML for in-database prediction. But, to make this work it seems like you need to use the PMML plugin by Zementis.  That means PMML is truly not open source.

Is there an easier option so that I can map PMML to SQL for prediction?

I have also looked for an option i.e. to use the JSON instead of XML to output model predictions.

But I'm confused where would R sit in such model as I'm assuming it's always needed to map to SQL.

But in this case, where would the R model sit? I'm assuming it would always need to be mapped to SQL.

Any other options out there?

Answer 1

Well, I could say that the answer completely depends on what production environment you are using

You can try out this new open source PMML "scoring  engine" called Pattern, only if you are using Big Data on Hadoop

Else your only choice is to run R on server. You will save your fitted models in .RData files and then load them to run respective predict on the server.

This would be very slow but you can always try to use more hardware.

Now, how you do that completely depends on the platform. You can always use user defined function to add custom function in R.

Coming to Hadoop you can add such functions to Pig or you can use RHadoop to write simple map-reduce code to load the model and call predict in R.

If your data are in Hive, you can use Hive TRANSFORM to call external R script.

There are also other ways such as vendor-specific to add functions written in R to various SQL databases.

Comments

RPubs.com.

you can do even dashboard and public.

---

Hey @Irina, Can you please expand your answer and explain a little more?