Subscribe Bookmark RSS Feed

JMP for Machine Learning

mikemalloy

New Contributor

Joined:

Jul 6, 2017

I would like to use JMP to speed up data analysis and creation of neural network prediction models.  Our runtime environment would be using the Google Cloud Platform and TensorFlow.  Can anyone point me to some good articles or videos about how to make the best use of JMP for that?  My business must be excellent at creating models for customers quickly and efficiently.  Would JMP be a tool that can make a difference in this?

4 REPLIES
txnelson

Super User

Joined:

Jun 22, 2012

1.  There is a good Document on the Machine Learning techniques available in JMP.  "Predictive and Specialized Modeling.  It is found at:

     Help==>Books==>Predictive and Specialized Modeling

2.  There are also a few videos available on Youtube.com.  Just search on "JMP Neural".

Jim
Peter_Bartell

Joined:

Jun 5, 2014

If you are serious about building and evaluating predictive models from large, perhaps messy, or unruly data sets, then seriously consider JMP Pro rather than standard JMP. JMP Pro has a much richer set of modeling methods as well as ancillary tools/utilities like very flexible model cross validation techniques, a Formula Depot for creating scoring code models in other languages such as C, Python, SAS or SQL, and model comparison tools as well.

ian_jmp

Staff

Joined:

Jun 23, 2011

Thanks to the work that Nascif and Dan put into this Discovery paper, I've found it to be a very useful resource retrospectively.

nascif_jmp

Staff

Joined:

Jul 30, 2015

As it was mentioned above, you will need JMP Pro to take advantage of JMP ability to generate scoring code for deployment. We do support Python code generation and have successfully deployed a JMP model in Spark. That would allow you to scale up to large data volumes. It would work in the Google Cloud Platform using their Google Dataproc offering.

 

Another deployment option would be to use our JavaScript scoring code generation capability and publish your neural network as a Serverless Google Function. You would still be able to scale up while having a much simpler (and cheaper) operation model. If you are coming to our next JMP Discovery in St Louis I would be happy to discuss this idea with you. I might even demo something similar. We will see. :)

 

 

I have looked into TensorFlow and I believe we could eventually support it as a code generation target, but it is not on our roadmap yet.