Linear Fit

import numpy as np
from smpl import plot
from smpl import io
from smpl import fit
from smpl import functions as f
import uncertainties.unumpy as unp
import uncertainties as unc

data = np.loadtxt(io.find_file('test_linear_data.txt',3))
xdata = data[:,0]
xerr = data[:,2]
ydata = data[:,1]
yerr = data[:,3]
x = unp.uarray(xdata,xerr)
y = unp.uarray(ydata,yerr)

data

SciPy

ff = plot.fit(xdata, ydata, fmt='.', label='data', xaxis="x in a.u.",yaxis="y in a.u.",function=f.linear, params=[1])

ff = plot.fit(xdata, ydata, fmt='.', label='data', xaxis="x in a.u.",yaxis="y in a.u.",function=f.line, params=[1,2])

ff = plot.fit(x, y, fmt='.', function=f.line, params=[1,1], sigmas=1,lpos=2)

ff = plot.fit(xdata, y, fmt='.', function=f.line, params=[1,1], sigmas=1,lpos=2)
print("Chi2 = ",fit.Chi2(xdata,y,f.line,ff))

Correlations

Let’s take a look at correlations on the example of calculating the y axis intercept. Again:

a,b = plot.fit(xdata, ydata, fmt='.', label='data', xaxis="x in a.u.",yaxis="y in a.u.",function=f.line, sigmas=1,extrapolate_min=-2)

Let’s calculated the intersection with the x axis, given by \(-b/a\), with correlation

-b/a

and without correlations

a_nocorr=unc.ufloat(plot.unv(a),plot.usd(a))
b_nocorr=unc.ufloat(plot.unv(b),plot.usd(b))
-b_nocorr/a_nocorr

While the mean agrees the uncertainties are not the same.

The Lines drawn in the plot are also correlated and cross the axis at -0.67 and -1.51 corresponding to an uncertainties of -0.42.

Minuit

ff = plot.fit(xdata, y, fmt='.', function=f.line, params=[1,1], sigmas=1,lpos=2,fitter=fit.Fitter.MINUIT_LEASTSQUARES)
print("Chi2 = ",fit.Chi2(xdata,y,f.line,ff))