0
$\begingroup$

I have an ensemble of data points $(x_i,y_i)$ and I need to fit them according to $y=F(a,b,x)$ if $x \leq x_{ref}$ and $y=G(c,d,x)$ if $x \geq x_{ref}$. The parameters I need to adjust are then $a,b,c,d,x_{ref}$. Models $F$ and $G$ are nonlinear with respect to the parameters.

For sure, if I select $x_{ref}$ among the $x_i$, I can run separate curve fits, add the sum of squares and change $i$ until I find a minimum.

Is there any way

  1. to automate this procedure with Mathematica (this would provide the best values of $a,b,c,d$ for the best $x_{ref}$ selected among the $x_i$)
  2. perform in a second step the full regression

Any help will be appreciated.

$\endgroup$
1
  • 1
    $\begingroup$ Please include a small complete example, it will help to work on something concrete. $\endgroup$ Commented Mar 16, 2014 at 10:06

1 Answer 1

3
$\begingroup$

If I understand your query, you want to do something like this (xref of 5 hard-coded here):

ClearAll[data, fitfn, x, a, b, c, d]

(* fake some non-linear data *)
data = {#, If[# < 5, 2 # ^1.5, 4 #^2.5]} & /@ Range[20];

(* fit function *)
fitfn[x_, a_, b_, c_, d_] := Piecewise[{{a x^b, x < 5}, {c x^d, x >= 5}}]

(* find parameters*)
sol = FindFit[data, fitfn[x, a, b, c, d], {a, b, c, d}, x];

(* data vs fit *)
Column[{sol,
  ListPlot[{Sort@data, Table[fitfn[x, a, b, c, d] /. sol, {x, 1, 20}]}, Joined -> True, 
   ImageSize -> 300]}, Left, 2]

enter image description here

$\endgroup$
3
  • $\begingroup$ This is a almost what I am looking for but it does not answer the problem of finding the $5$ which is a parameter to be adjusted. The function must have the $5$ as a tunable parameter. Would you find a way to answer the first question (loop around the $x_i=x_{ref}$ to find the best point) and complete the second one with $x_{ref}$ as adjustable parameter). Thank you ! $\endgroup$ Commented Mar 16, 2014 at 12:39
  • $\begingroup$ @ClaudeLeibovici: No. Everything you need to polish this and add search is in the answer, this is not a code-for-free site. Make an attempt and show your code, if you get stuck or have problems, post a new question or update this one. $\endgroup$
    – ciao
    Commented Mar 18, 2014 at 8:36
  • $\begingroup$ I am not looking for a code-for-free site but for help since I am a more than limited user. $\endgroup$ Commented Mar 18, 2014 at 8:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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