cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Try the Materials Informatics Toolkit, which is designed to easily handle SMILES data. This and other helpful add-ins are available in the JMP® Marketplace
Choose Language Hide Translation Bar
ivomdb
Level III

reporting R-square for customised nonlinear model fit

I am trying to find the best way to report a nonlinear model fit to my experimental data. 

 

I have managed to write a script that fits a nonlinear model and finds the parameters estimates using the loss function. It then reports each parameter estimates and the non linear errors.

However, I would like to get the RMSE or R-square of the resulting fit also. Is this not available for nonlinear models? or do I need to add an extra analysis step?

 

When running the nonlinear analysis platform interactively the resulting plot shows both the experimental data and the model curve. However, when running the script automatically only the experimental data is shown. Does anyone know why this happens? Am I missing anything on the script below?

 

Could someone provide some literature/books/links to be able to get the graphs of the data vs resulting model fit and report the R-square of the fit?

 

 

The script I am using for the nonlinear model fit is the one below at the moment:

 

col1 = New Column( "Model" );
col1 << set formula(
	Parameter(
		{Y1_Y1_0 = 0.2, K1_K1_0 = 0.01, P1_P1_0 = 0.004, Y2_Y2_0 = 0.15, K2_K2_0 = 0.06, P2_P2_0 = 0.3},
		If( :Days > 0,
			Y1_Y1_0 * Exp( -Exp( ((K1_K1_0 * Exp()) / Y1_Y1_0) * (P1_P1_0 - :Days) ) ) + Y2_Y2_0 *
			Exp( -Exp( ((K2_K2_0 * Exp()) / Y2_Y2_0) * (P2_P2_0 - :Days) ) ),
			0
		)
	)
);
col2 = New Column( "Err" );
col2 << set Formula( Sum( (:Average - :Model) ^ 2 ) );
col2 << Evalformula;
col2 << Get formula;
obj = Nonlinear( Y( :Methane Yield ), X( :Model ), Loss( :Err ), Iteration Limit( 600 ), By( :Sample ) );
obj << go;
//returns a list of vectors where each vector, contains the parameters estimates
obj << get estimates;
//also see Get SSE, Get Parameter Names, Get CI
//Alternately, make a table
_xx = obj << Xpath( "//OutlineBox[@helpKey='Nonlinear Solution']" );
//only 2 groups so only 2 are found; each contains two sub tables
fit_dt = (_xx[1] << Find( Table Box( 1 ) )) << Make Combined Data Table;
est_dt = (_xx[1] << Find( Table Box( 2 ) )) << Make Combined Data Table;
fit_dt << set Name( "NonLin Fit Errors" );
est_dt << set Name( "NonLin Fit Estimates" );

 

 

 

 

 

 

 

 

3 REPLIES 3

Re: reporting R-square for customised nonlinear model fit

The RMSE is reported in the Solution outline:

 

Capture.PNG

 

The R square is not available because you need two models: the fitted model and the reduced model. There is no reduced model in the case of the non-linear model. (The reduced model in the OLS regression is the model with only the constant intercept term.)

 

You are incorrectly re-defining the default loss function in Nonlinear. The default loss is least squared error. It appears from your script that you want to use least squared error. Your column formula should result in the loss for each observation, not for the entire sample. So your loss formula should only be the squared difference. (That is, do not compute the sum of the squared differences.) With this approach instead of the default loss, I get:

 

Capture.PNG

 

So the use of a custom loss changes the information in the Solution outline to the total Loss (SSE in this case), the Avg Loss (MSE in this case), and the Sqrt Avg Loss (RMSE in this case).

 

I saved the script from Nonlinear and ran it. The data plot still shows the curve representing the graph of the nonlinear function.

ivomdb
Level III

Re: reporting R-square for customised nonlinear model fit

Thanks Mark.

 

I have changed the Loss function as you suggested. I have run the fitting but still can't see the model curve on the Plot.

 

On my Solution I get: Avg Loss and Sqrt Avg Loss. Does this mean my RSME is the avaerage loss then?

Re: reporting R-square for customised nonlinear model fit

Your RMSE is the Sqrt Avg Loss.

 

I don't know the cause of the absent plot.