CODEWITHSUNDEEP
pythonscipyscipy-optimize
Curo1305
Curo1305

Reputation: 21

Is there a way to have multiple variables fixed and fitted for the SciPy curve_fit function?

I have a problem with the curve_fit function of scipy.optimize. I have a rather complex class which can calculate the propagation of light and I want it to be fitted to some measurements. For this thread, I will give a simple example of the structure of this class

import numpy as np
from scipy.optimize import curve_fit
import matplotlib.pyplot as plt

class Gauss():
    
    def __init__(self,x,parDict={}):
        self.default_dict = {"Amplitude": 1, "Center": 0, "FWHM":1} 
        self.x = x
        if parDict:
            self.parDict = self.fillDict(parDict)
        else:
            self.parDict = self.default_dict
    
    def fillDict(self,dictionary):
        return_dict={}
        if "Amplitude" in dictionary.keys():
            return_dict["Amplitude"] = dictionary["Amplitude"]
        else:
            return_dict["Amplitude"] =  self.default_dict["Amplitude"]
        if "Center" in dictionary.keys():
            return_dict["Center"] = dictionary["Center"]
        else:
            return_dict["Center"] =  self.default_dict["Center"]
        if "FWHM" in dictionary.keys():
            return_dict["FWHM"] = dictionary["FWHM"]
        else:
            return_dict["FWHM"] =  self.default_dict["FWHM"]
        
        return return_dict
    
    def calculate(self):
        return self.parDict["Amplitude"]*np.exp(-((self.x-self.parDict["Center"])**2/(self.parDict["FWHM"]**2*np.sqrt(2))))

x = np.linspace(-5,5,100)       
g = Gauss(x)
y = g.calculate()

plt.plot(x,y)
plt.show()

So one can see, I have all the parameters inside a dictionary for simpler use and a default parameter dictionary for all the missing parameters I don't want to change. Now I am struggling to find a clever way or even any way to write a function or class which can use this class to fit it to some measurements. Like:

ynoise = y + np.random.normal(loc=0.2, scale=0.2, size=len(x))
param = {"Amplitude":1.2,"FWHM":0.5}
fit_par,_ = curve_fit(g.calculate, x, ynoise , p0=param)

The problem is, that this class is part of a much larger programming project and inside a program so I don't have the meaning of manipulating the class itself. Can one of you think of a solution?

Upvotes: 1

Views: 219

Answers (1)

Curo1305
Curo1305

Reputation: 21

With the comment by Bob with his answer to this post, I could create a solution for my problem.

For everyone who want the solution:

def fitFunc(x, dictionary):
    g = Gauss(x,dictionary)
    return g.calculate()

fit_par,_ = curve_fit(lambda x,*args: fitFunc(x, {k:v for k,v in zip(param.keys(),args)}),x, ynoise, [param[k] for k in param.keys()])
fit_par = {k:v for k,v in zip(param.keys(), fit_par)}

Thank you very much!

Upvotes: 1

Related Questions