CODEWITHSUNDEEP
pythonnumpyscipycurve-fittinggaussian
icypy
icypy

Reputation: 3202

My use of Scipy curve_fit does not seem to work well

I am trying to to fit my data to a Gaussian using Scipy curve_fit. Unfortunately curve_fit returns 1,1,1. 

Kp4=fnamer4[615:645]
xk=np.arange(0,1024,1)[615:645]

def func(x, a, x0, sigma):
return a*np.exp(-(x-x0)**2/(2*sigma**2))
popt, pcov = curve_fit(func, xk, Kp4)
print (popt)
Kp4_fit= func(xk, popt[0], popt[1], popt[2])
plt.plot(xk, Kp4_fit, 'r',xk, Kp4, 'bs')

Kp4 is equal to

>>> Kp4
array([23, 27, 20, 26, 22, 22, 26, 29, 32, 19, 34, 26, 29, 24, 32, 41, 27,
   39, 33, 30, 30, 30, 26, 39, 30, 21, 17, 16, 17, 14])

The output of the print popt is [ 1. 1. 1.] I have tried curve_fit on much different data before, and it worked fine. Perhaps the problem is that curve_fit cannot fit the data the Gaussian function?! Thanks for any help.

The concept I am using for the curve_fit is based on an example from a link bellow: http://python4esac.github.io/fitting/examples1d.html

Upvotes: 1

Views: 176

Answers (1)

jkalden
jkalden

Reputation: 1588

You have to pass an initial guess for popt, otherwise the fit starts with [1,1,1] as initial guess which is pretty poor for your dataset!

The following gives reasonable results for me:

popt, pcov = curve_fit(func, xk, Kp4, p0=[20,630,5])

The initial guess could be [np.mean(Kp4), np.mean(xk),5*(max(xk)-min(xk))/len(xk)], to have a general starting point.

Don't forget to put plt.show() or something similar at the end.

Upvotes: 1

Related Questions