What should I do if after I run the nonlinear regression and the following message appears on the control panel: "Consider looser convergence criteria or selec "accept current estimates" under "iteration options". Failed: Cannot decrease objective function

[#1]

The code is working good for other samples, but for few samples it is showing that the graph is not convergent.

"Consider looser convergence criteria or select "accept current estimates" under "iteration options". Failed: Cannot decrease objective function

The code is as below and I am attaching the samples for which it is not working well.

Names Default To Here( 1 );

dt= currentdatatable();
New Column( "Model",
	Numeric,
	"Continuous",
	Format( "Best", 12 ),
	Formula(
		Parameter(
			{p = 0.29, x0 = 15, m0 = 1.2, x1 = 30, m1 = 5},
			p * (1 - Exp( -(X_mm / x0) ^ m0 )) + (1 - p) * (1
			-Exp( -(X_mm / x1) ^ m1 ))
		)	
	),

);


nlin = dt << Nonlinear(
SendToByGroup( Bygroup Default ),
Y( Ln_C ),
X( :Model ),
Iteration Limit( 1000 ),
Unthreaded( 1 ),
QuasiNewton BFGS,
//Newton,
Finish,
By( Sample_ID ),

);

[#2 · Accepted solution]

Thank you for providing the example. It is beneficial when diagnosing the problem.

I interactively launched Nonlinear and fit the model for each sample. I see the same notice. It is based on the criteria you used to determine convergence. None of the criteria were satisfied. On the other hand, you have a good fit! The SSE is relatively small.

I do not think it is a data problem or an inappropriate model. In fact, I clicked Reset and then Go. It converged by the Gradient criterion the second time for every sample.

[#3]

Thank you sir, by clicking on Reset and Go, the data is being converged by Gradient criterion. But, what if I wanted to automize it in the script instead of resetting the graph because I wanted to use this output for further analysis. And only these samples are not being converged in the first attempt, so how to automize this using the script.

[#4 · Accepted solution]

if it seems that a failure to converge requires a single Reset and Go, then your script could send the messages in succession to each sample. See the Scripting Index for Reset and Go messages.