488025_en-US

liliconcom

Add Isolation Forest as an approach for Unsupervised Learning

Add Isolation Forest as an approach for Unsupervised Learning

Statistics and AnalyticsRe: Add Isolation Forest as an approach for Unsupervised Learning

We started work to call python code from JMP using Isolation Forest as our first test case and have had some initial success. I am planning to set it up as an add-in.

I still would prefer to see it a an Anomaly detection option in JMP.

Re: Add Isolation Forest as an approach for Unsupervised Learning

Thanks Peter

Re: Add Isolation Forest as an approach for Unsupervised Learning

Here is a good place to start.  https://www.jmp.com/en_us/whitepapers/jmp/jmp-synergies-using-jmp-and-jmp-pro-with-python-and-r.html

 

Re: Add Isolation Forest as an approach for Unsupervised Learning

I would not mind testing/creating an add-in that uses the Isolation Forest python call in Sklearn or the one in Pyod.

However. I am not proficient enough in the JMP/Python interface to create the add-in. Can somebody contact me to provide guidance.

Re: Add Isolation Forest as an approach for Unsupervised Learning - Status changed to: Investigating

We are investigating this for a possible new feature, but there are no definite plans right now to put this into JMP.   It might be interesting to prototype this with a JMP addin that calls an open source tools, such as the isotree R package.  

Re: Add Isolation Forest as an approach for Unsupervised Learning

Isolation Forest is an Unsupervised Learning approach for Multivariate Anomaly Detection.

It would be used in a similar way to the Mahalanobis or T2 distance approach for outliers, although it uses non-parametric (tree based) approach

 

Re: Add Isolation Forest as an approach for Unsupervised Learning - Status changed to: Needs Info

@rosen0 this is an interesting idea.  Can you share what kind of problem you would use this for?  We want to understand the use case better.  

Re: Add Isolation Forest as an approach for Unsupervised Learning

First... IsolationForest fits nicely since JMP already has Boosted trees and random Forest for supervised learning

 

Both, from literature and some experience at P&G, IsolationForest are fast and effective. Because they are Tree based it is also possible to get efficiently Importance Contributions 

Re: Add Isolation Forest as an approach for Unsupervised Learning

I am curious to know how differently isolation forest work compared to the classifcal outliers analysis algorithm 

Re: Add Isolation Forest as an approach for Unsupervised Learning - Status changed to: Not Planned For Now

We currently do not plan to build this into the product, but we are hoping this can be delivered via an add-in utilizing the new python integration, as discussed in the comments. 

Re: Add Isolation Forest as an approach for Unsupervised Learning

We now have a working add-in for Isolation Forest using python integration.