Splunk ML toolkit export Predictive models and Realtime scoring

[thomas13t]

Hello, I have a couple of questions regarding Splunk ML toolkit:

1. Can I export Predictive models / algorithms from Splunk? If so what are the available exports: can it be code in python? can it be in R?

2. The algorithms will need to run on (near)realtime data flows, can Splunk handle real time Predictive scoring? A scoring engine, something like SAS ESP (event stream processing), that can deployed separately form the Splunk instance?

Thx
Thomas

[skoelpin · Accepted solution]

Hello,

1) Splunk uses the Python Scientific Compute add-on for its algorithms, so no, you cant export it, but you can use the Python based scientific compute kit

https://docs.splunk.com/Documentation/MLApp/3.1.0/User/Installandconfigure

2) It depends on a lot of things. How you set up your models, your hardware, your sample size, how many independent variables you have etc.. It's possible, but it depends

[thomas13t]

Ok Thank you,

So if I understood correctly it could be theoretical possible to decouple the analytic part (creation, test, and validation of predictive models) from a "scoring" engine where the models are deployed, the needed flow (and eventual extra KPis are calculated) are present?

Or with your 2 answer you are implying that the splunk instance will be called via API everytime a resource needs a scoring and reply with the correct score?