Deploying models

KNIME Analytics Platform
1

Hello,

I have a hard time finding the answer to a basic question:

Once you are fairly satisfied with your model (e.g a classifier), how do you go about and put it to work?

Some solutions (e.g. SAS) export a model to code (C, java,...) so you can actually incorporate the model in your own system, using the tool/language of your choice. Most of the others I've seen so far, force you to use their own server-product. Azure-ML (in preview) has a button that creates an Azure webservice from your model, with the click of a button. SPSS Modeler forces you to buy their scheduling server.

So how about Knime? So far, I've only found one post in this forum that basically says: please contact us to buy Knime Server. Are there any alternatives?

For example:

  • run the Knime model on a command-line (it's a Java program, so why not...).
  • use a Knime library/module/package for java/C#/python/....
  • exporting the model to ... (not entirely sure what to put here :) )
  • something else entirely

I'm really curious to find out how people deploy/use the models they build with Knime.

Tim

2

Hi Tim,

A few options come to mind:

  • Use their server product :)
  • Use the command-line interface to trigger workfows remotely (someone has even wrapped this into PHP with a Bootstrap interface recently - can be found on the forum)
  • Use PMML and deploy to Zementis or any other PMML-compatible execution engine
  • Somehow convert your model to to
    • SQL (DYMATRIX has a commercial PMML node for that)
    • Java IF-statements (for decision trees)
    • R to use Rserve or similar
    • Groovy
    • Octave
    • MatLAB
    • ...

Admittedly using Java "as is" would be nice. Not sure if this would be a freebie, though. Apart from the server, it'd probably be the only way to support all of KNIME's models.

Cheers
E

3

Also depends what models. AFAIk R and Weka models can be exported and then used with R or Weka outside of KNIME.