57156_en-US

liliconcom

Automating custom nonlinear regression model

Hi all,

 

I'm trying to automate a nonlinear regression routine and part of that involves creating a custom model via column formulas. 

 

I want to have an arbitrary number of parameters, and I think I've solved that by dumping an appropriately formatted list into the Parameter({}) section ("Parm_Start").  One problem I'm having is that I don't know how to arbitrarily set the term in Match().  I tried to do this with Match_List{}, which matches the format, but it gives odd outputs.

 

Secondly, I'm trying to sum a series of exponentials with "o" pre-exponential terms and "o" using the form A[o]*exp(-t/t[o]) -- see the code below.  The A's are all fittable parameters, pulled from a column Mod_Parm; Time is a column of arbitrary length and the independent variable; and Timeconst is a column of the same length as Mod_Parm (both of which are shorter than Time).

 

I haven't gotten to the part where the JSL-based nonlinear modeling routine is called because need to fix this first.

 

Thanks!

 

 

Mod_Parm_list={};
Parm_start={};
Parm_list={};
Match_list={};
b=1;

for(m=1,m<=num_exp,m++,
 Mod_Parm_list[m]="A"||char(m);
 Parm_start[m]=parse("Parm_"||Mod_Parm_list[m]||" = -0.1");
 Parm_list[m]="Parm_"||Mod_Parm_list[m];
 Match_list[b]=Mod_Parm_list[m];
 Match_list[b+1]=parse(Parm_list[m]);
 b=b+2;
);

dt << NewColumn("I_model",Format ("Scientific"),Set Property( "Expand Formula", 1 ),
 Formula
  (Parameter(
   {Parm_start},
   I_DC - Summation(o=1,num_exp,
   Match(:Mod_Parm,
    "A1", Parm_A1,
    "A2", Parm_A2,
    "A3", Parm_A3,
    "A4", Parm_A4,
    "A5", Parm_A5,
    "A6", Parm_A6    
    )*Exp(-:Time/:TimeConst[o])
    )
   )
  )
 );

 

Re: Automating custom nonlinear regression model

For posterity, I wanted to add that part of the solution can be found here:

https://community.jmp.com/t5/Discussions/How-to-pass-a-variable-into-a-formula/td-p/52344

 

That is, part of my issue was not defining "I_DC" as a table variable and assuming the column formula would understand the expression.  Here's the update for that part:

I_DC = 3; //Using only this value in the column formula is wrong
dt = Current Data Table();
dt << set Table Variable( "I_DC_form", I_DC ); //Allows the parameter to be used later in the col formula expression

Mod_Parm_list={};
Parm_start={};
Parm_list={};
Match_list={};
b=1;

for(m=1,m<=num_exp,m++,
 Mod_Parm_list[m]="A"||char(m);
 Parm_start[m]=parse("Parm_"||Mod_Parm_list[m]||" = -0.1");
 Parm_list[m]="Parm_"||Mod_Parm_list[m];
 Match_list[b]=Mod_Parm_list[m];
 Match_list[b+1]=parse(Parm_list[m]);
 b=b+2;
);

dt << NewColumn("I_model",Format ("Scientific"),Formula(
  Parameter(
   {T1 = 1, T2 = 1, T3 = 1, T4 = 1, T5 = 1, T6=1},
   :I_DC_form + T1 * Exp( -:Time / 0.004 ) + T2 * Exp( -:Time / 0.01 ) + T3 * Exp( -:Time / 0.1 )
    + T4 * Exp( -:Time / 1 ) + T5 * Exp( -:Time / 10 ) + T6 * Exp( -:Time / 100 ))
   )
  );