For some FittedModel, the "BestFitParameters" are given in terms of the symbols used to define the model.

fit = NonlinearModelFit[{10,11,12},a*x+c,{a,c},x];

returns {a->1.,c->9.}

This can be problematic if I define a or c somewhere else. One option is to use the model in a module, and try to localize variables but this generates unique keys that I have to keep track of:

fit=Module[ {a,c}, NonlinearModelFit[{10,11,12},a*x+c,{a,c},x] ];

returns {a$768206 -> 1., c$768206 -> 9.}

Often I'll do something like

fit=Module[ {a,c,lf},  lf=NonlinearModelFit[{10,11,12},a*x+c,{a,c},x]; {"a"->a,"c"->c}/lf];

which returns {"a"->1.,"c"->9.} with strings as keys.

I actually prefer this since now "a" and "c" can't clash with a definition of the symbols a or c, but it's a pain because I lose access to the FittedModel object which may be useful later on.

Since I can't use the strings "a" or "c" in the NonlinearModelFit function, my question is this: Is there a way to modify the FittedModel object, such that requesting "BestFitParameters" returns the a list of rules with strings as the keys?

Alternatively, does anyone have a more elegant way of working with these objects, so I don't have to keep track of what symbols I use in fits and make sure not to use the same symbols to define similar values elsewhere?


Why not use formal symbols, e.g., \[FormalA] or \[FormalC]? These symbols are protected, so they should never acquire a value:

fit = NonlinearModelFit[
    \[FormalA] x + \[FormalC], {\[FormalA], \[FormalC]}, x

enter image description here

  • $\begingroup$ I didn't even know these existed! Thanks for the option. $\endgroup$ – N.J.Evans Mar 29 at 12:11
  • $\begingroup$ Just a note, you can get a normal characters with esc-$a-esc or \[FormalA], greeks are available too, but you have to spell out the letter, esc-$Beta-esc or \[FormalBeta] - you can't use the character. $\endgroup$ – N.J.Evans Mar 29 at 12:56

One straightfoward way to handle this is to accept the unique keys generated inside the module and write a function that replaces these with de-unique-ified strings when the best fit parameters are needed:

getBestFit[fit_FittedModel] := Module[
   {a, c, bf, newKeys, x, oldkeys},
   bf = fit["BestFitParameters"];
   oldkeys = Keys@bf;
   newKeys = First /@ StringSplit[ToString /@ oldkeys, "$"];
   Rule @@@ Transpose@{newKeys, oldkeys /. bf}

fit = Module[
   {a, c, x},
   NonlinearModelFit[{10, 11, 12}, a*x + c, {a, c}, x]

shows that the unique symbols generated inside the module are retained

{a$772530 -> 1., c$772530 -> 9.}



returns the desired rules with strings as keys


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.