Gauss-Newton

Copyright (C) 2020 Andreas Kloeckner

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

import numpy as np
import scipy as sp
import matplotlib.pyplot as pt
import scipy.linalg as la

We would like to fit the model $f(t) = x_0 e^{x_1 t}$ to the following data using Gauss-Newton:

t = np.array([0.0, 1.0, 2.0, 3.0])
y = np.array([2.0, 0.7, 0.3, 0.1])

First, define a residual function (as a function of $\mathbf x=(x_0, x_1)$)

def residual(x):
    return y - x[0] * np.exp(x[1] * t)

Next, define its Jacobian matrix:

def jacobian(x):
    return np.array([
        -np.exp(x[1] * t),
        -x[0] * t * np.exp(x[1] * t)
        ]).T

Here are two initial guesses. Try both:

#x = np.array([1, 0])
x = np.array([0.4, 2])

Here's a plotting function to judge the quality of our guess:

def plot_guess(x):
    pt.plot(t, y, 'ro', markersize=20, clip_on=False)
    T = np.linspace(t.min(), t.max(), 100)
    Y = x[0] * np.exp(x[1] * T)
    pt.plot(T, Y, 'b-')
    
    print("Residual norm:", la.norm(residual(x), 2))

plot_guess(x)

Residual norm: 162.72704101675862

Code up one iteration of Gauss-Newton. Use numpy.linalg.lstsq() to solve the least-squares problem, noting that that function returns a tuple--the first entry of which is the desired solution.

Also print the residual norm. Use plot_guess to visualize the current guess.

Then evaluate this cell in-place many times (Ctrl-Enter):

x = x + la.lstsq(jacobian(x), -residual(x))[0]

plot_guess(x)

Residual norm: 0.044677532987197045