Scipy curve_fit with a parameters matrix of unknown shape

Question

I am trying to use either scipy.curve_fit or scipy.least_squares for a function like

def f(x, C): 
    
  r = 0
  for i in range( len(C) ) :
      for j in range( len(C[i]) ):
          r+= x[0]**j * x[1]**i * C[i][j]          
  return r

where C is a list of lists of length defined run time. One of the parameters of the code is something like [[1., 1.]] or [[1., 1.], [1.]] which defines the shape of C.

I tried following this answer, but there the C would be a NxN matrix, and curve_fit does not work well with matrices as input. I tried to define a function with x, *args as parameters but with no luck for now.

I also tried following this other answer which uses least_squares directly, but also here I get the obstacle that I do not know the size of the rows beforehand.

I feel that a combination of the two answers can work for me but I cannot figure it out. I was thinking to write a function that has as parameters the length of C, a list of the lengths of C[i], and then the elements of C flattened, but I cannot write a working version of the code. If anyone has a clear understanding of how to do that, or had a similar problem, I'd be thankful for any help.

Not necessarily, it is the initial guess. In my code it serves to give the shape mostly, how many rows and how many columns per row — Bagnarol, Jul 20 '23 at 12:46

score 0 · Answer 1 · answered Jul 20 '23 at 12:45

0

Since this question got some recent activity I will answer it, as I solved it few days later. I solved it with a lambda function.

I defined my function as

def f(xy, *args, dimC = 1, dimCi = [2]): 

   x, y = xy # I have to unpack the tuple because a normal list seems to not work

   r = 0

   for i in range( dimC ) :
       for j in range( dimCi[i] ):
           r+= args[i*dimC + j] * x**i * k2**j
        
   return r

Then I just need to flatten the initial guess and I am free to call

C, cov_C = curve_fit(lambda x, *p: f(x, *p, dimC = dimMat, dimCi = dimMati), ( x, y ), z), p0=init_guess)

answered Jul 20 '23 at 12:45

Bagnarol

13
6

We don't have `dimMat` or `dimMati`, among other things; so this answer is not reproducible. – Reinderien Jul 20 '23 at 12:54
As it’s currently written, your answer is unclear. Please [edit] to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers [in the help center](/help/how-to-answer). – Community Jul 23 '23 at 05:14
@Reinderien Just read the question and then the answer. The entire point is that you have a list of lists of unequal length that has to be defined at runtime. dimMat and dimMati are the number of rows and the columns per each rows (which, in general, varies). Of course you need to adapt it to your code. This is clear if you compare f in the answer with f in the question. – Bagnarol Jul 24 '23 at 07:31

Scipy curve_fit with a parameters matrix of unknown shape

1 Answers1