CODEWITHSUNDEEP
pythoncurve-fittingcovarianceerrorbarlmfit
lAPPYc
lAPPYc

Reputation: 1

How do I get uncertainties from lmfit minimizer

I'm using lmfit to fit a bunch of lorentzians. For reasons of boundaries, I can't use scipy.optimize functions directly. With lmfit the fits work well, but res.covar is always None, res.errorbars is False, res.uvars is undefined and res.params['parameter_name'].stderr is also None. I need the errorbars. Here's my code (without data):

def to_minimize(params, x, y, uncertainties=None):
    x0 = params["x0"]
    ampl0 = params["ampl0"]
    ampl = params["ampl"]
    sig = params["sig"]

    model = lorentzian(x, x0, ampl0, ampl, sig)
    if uncertainties is not None:
        return (y- model) / uncertainties
    else:
        return y - model


params = lmfit.Parameters()
params.add("x0", value=guess[0], min=guess[0] - 1000, max=guess[0] + 1000)
params.add("ampl0", value=guess[1], min=0)
params.add("ampl", value=guess[2], min=0)
params.add("sig", value=guess[3], min=0)

out = lmfit.minimize(to_minimize, params, args=(x_data, y_data))

Here, lorentzian is a function that calculates the lorentzian and guess is a list of initial values for the guess. x_data and y_data are 1d arrays of 50000+ size. Here's the output that I get when I type out in a jupyter cell:

Fit Result
Fit Statistics
fitting method  leastsq
# function evals    156
# data points   51201
# variables 4
chi-square  9.8949e-23
reduced chi-square  1.9327e-27
Akaike info crit.   -3149414.19
Bayesian info crit. -3149378.82
Parameters
name    value   initial value   min max vary
x0  2424682.92  2424725.9731872166  2423725.97  2425725.97  True
ampl0   1.6720e-13  1.6717888361052836e-13  0.00000000  inf True
ampl    1.2557e-12  1.2557626695414776e-12  0.00000000  inf True
sig 105.300229  4.296879768371582   0.00000000  inf True

print(out.covar) gives None. out.errorbars is False. out.uvars raises AttributeError. I need these errors in the parameters, no matter how small. But I get no warnings during the fit, and numdiftools is installed. What am I missing here?

My lmfit version is 1.3.2 and numdifftools is 0.9.41. Python is 3.11.7

If it's impossible to get them as outputs, can you suggest a way to calculate these by hand? Thanks a lot

Upvotes: 0

Views: 40

Answers (0)

Related Questions