BookmarkSubscribe
Choose Language Hide Translation Bar
S_Xue
Community Trekker

How to run standard least square regression 1000 times with different validations?

How to run standard least square regression 1000 times with different validations (bootstrap: 1000 validation and training groups)? In the results file, there are RSquares for different groups and 1000 iterations, and the factor profiling results. Thank you.

 

Dear all,

I am a new learner of JMP Pro. I got the data of 7 different predictors and one response factor. I want to do the standard least square regression 1000 times with different validations. To compare this analysis with neural networks, this can do this easily with the script Validation Method („Holdback“, 0.1). 

Now I can get the RSquares successfully. But I still do not know how to get the factor profiling (the importance of each predictor) results for different validations.

 

Previously, I was replied from a kind JMP staff that I can use this script to get RSquares for different validations in Least Square Regression. (https://community.jmp.com/t5/Discussions/How-to-make-different-validation-automatically-using-JMP-sc...)

dt<<Make Validation Column(
	Training Set( 0.90 ),
	Validation Set( 0.10 ),
	Test Set( 0.00 ),
	Formula Random

 

Thank you.   

0 Kudos
2 ACCEPTED SOLUTIONS

Accepted Solutions
Highlighted

Re: How to run standard least square regression 1000 times with different validations?

I think what you want to do (assuming you have JMP Pro) is just right click on the statistic that you want to bootstrap on (i.e. the RSquare the Cross-validation section of the Fit Model Report) and select Bootstrap. You can then choose the number of iterations and it will generate a data table and report of the results.
gzmorgan0
Super User

Re: How to run standard least square regression 1000 times with different validations?

Attached is a script written for Chapter 9 of JSL Companion, 2nd Edition.  It demonstrates:

  • a bootstrap of a 2 sample t-test from a Oneway analysis (only 100 was chosen)
  • how to get the results
  • how to plot the results
  • as an extra is shows how to order customizations by sending clicks to the customize framebox window.

image.png

4 REPLIES 4
Highlighted

Re: How to run standard least square regression 1000 times with different validations?

I think what you want to do (assuming you have JMP Pro) is just right click on the statistic that you want to bootstrap on (i.e. the RSquare the Cross-validation section of the Fit Model Report) and select Bootstrap. You can then choose the number of iterations and it will generate a data table and report of the results.
S_Xue
Community Trekker

Re: How to run standard least square regression 1000 times with different validations?

Thank you for your reply. Yes I am using the JMP Pro. Sorry that I did not find the statistic used for bootstrap.

 

 

Here is my script: I just want to run this analysis for 1000 times using 1000 different validations. Thank you.

Fit Model(
	Validation( :Validation ),
	Y( :Std meanDT ),
	Effects(
		:Std wl
	),
	Personality( "Standard Least Squares" ),
	Emphasis( "Effect Leverage" ),
	Run(
		Profiler(
			1,
			Confidence Intervals( 1 ),
			Dependent Resampled Inputs( 1 ),
			Arrange in Rows( 4 ),
			Reorder X Variables(
				:Std wl
			),
			Term Value(
				Std wl( 0, Lock( 0 ), Show( 1 ) )
			)
		),
		:Std meanDT << {Lack of Fit( 0 ), Plot Actual by Predicted( 1 ),
		Plot Regression( 0 ), Plot Residual by Predicted( 1 ),
		Plot Effect Leverage( 1 )}
	)
); 

 

0 Kudos

Re: How to run standard least square regression 1000 times with different validations?

I’m not aware of being able to bootstrap via a command in JSL. The way I was proposing was through the GUI. You might have a look at the example under the Resample Freq heading in the scripting index for an idea about how to set it up.

 

UPDATE - There is a JSL message function (search the scripting index for "bootstrap" to find it)... you learn something new every day!!  LOL.

M

gzmorgan0
Super User

Re: How to run standard least square regression 1000 times with different validations?

Attached is a script written for Chapter 9 of JSL Companion, 2nd Edition.  It demonstrates:

  • a bootstrap of a 2 sample t-test from a Oneway analysis (only 100 was chosen)
  • how to get the results
  • how to plot the results
  • as an extra is shows how to order customizations by sending clicks to the customize framebox window.

image.png