Ensemble Modeling in JMP

[#1]

Ensemble Models.mp4

Video Player is loading.

Play Video

Play

Mute

Current Time 0:00

/

Duration 4:22

Loaded: 0%

0:00

Stream Type LIVE

Seek to live, currently behind liveLIVE

Remaining Time -4:22

1x

Playback Rate

Chapters

• Chapters

Descriptions

• descriptions off, selected

Captions

• captions settings, opens captions settings dialog
• captions off, selected

Audio Track

• en (Main), selected

Picture-in-PictureFullscreen

This is a modal window.

Beginning of dialog window. Escape will cancel and close the window.

Text

ColorWhiteBlackRedGreenBlueYellowMagentaCyanTransparencyOpaqueSemi-Transparent

Background

ColorBlackWhiteRedGreenBlueYellowMagentaCyanTransparencyOpaqueSemi-TransparentTransparent

Window

ColorBlackWhiteRedGreenBlueYellowMagentaCyanTransparencyTransparentSemi-TransparentOpaque

Font Size

50%75%100%125%150%175%200%300%400%

Text Edge Style

NoneRaisedDepressedUniformDropshadow

Font Family

Proportional Sans-SerifMonospace Sans-SerifProportional SerifMonospace SerifCasualScriptSmall Caps

Reset restore all settings to the default valuesDone

Close Modal Dialog

End of dialog window.

Close Modal Dialog

This is a modal window. This modal can be closed by pressing the Escape key or activating the close button.

See how to  make ensemble models (models of models) in JMP 16 Pro using the Model Screening platform.  Ensemble models often are better predictors than individual models. 

Ensemble Model Set-Up Options

1. Build multiple models to choose the best models. Analyze>Predictive Models> Model Screening.
   1. Add Factors and Response and include Validation column.
   2. Choose and select Methods, Options, Folded consolidation, and Modeling Options.
   3. Click OK.
   4. Examine and compare results using R-square values in Evaluate Methods report.
   5. Pick several good models to build your ensemble model.
   6. From red triangle, Save Fast Formulas to add the model values to your data table.
2. Build ensemble model - Analyze>Predictive Model>Choose a Model and use Models and Predicted Y as Factors.  Also include Validation Column.
3. Compare R-square for ensemble model to individual model R-squares.  

Ensemble Model Results

[dale_lehman]

Nice, succinct introduction.  But I was wondering why you didn't use the Model Comparison option, which allows you to automatically do model averaging?  Instead, you built a neural net using the various model predictions.  Can you provide any guidance concerning whether averaging is (or is not) a good way to build an ensemble model?

[Peter_Hersh]

@dale_lehman,  Great question.  Model averaging is not exactly the same as Ensemble modeling.  The basic difference is Model averaging is taking an average of all of the model's predictions.  Each model has the same weight in the model averaging.  This typically leads to a more robust model and sometimes a better overall model, but not always.  I think this is a valid technique and worth using even in conjunction with ensemble modeling.  Ensemble modeling allows JMP to essentially apply a weight to more accurate predictions.  It is a bit more complex when you use multiple nodes and hyperbolic Tangent activation type in the Neural platform, but it is essentially what an Ensemble model does.

[gail_massari]

More about ensemble models