ICT 2102: Engineering

Computation
Sessional
1.5 Credits, 3 hrs/week

Mahin Sultana
Lecturer
Department of Civil Engineering
Bangladesh University of Business &
Technology
Curve-fitting and Interpolation
Curve-fitting and engineering practice
sampling data are acquired
at discrete points.
-but the values at undefined points
are wanted in many engineering
applications

Interpolation
- Ex: density of water at different
temp.

Least-squares Regression
-Purpose is trend analysis or
hypothesis testing
Trend of experimental data
Experimental data for force (N) and velocity (m/s) from a
wind tunnel experiment

v, m/s 10 20 30 40 50 60 70 80
F, N 25 70 380 550 610 1220 830 1450

1500

Is the relationship
1000
-linear? Which order?
(Linear or polynomial
F (N)

regression)
500

-nonlinear?
0
(Nonlinear regression)
0 20 40 60 80
v (m/s)
 Fitting the data with a straight line
Minimizing the sum of squares of the estimated residuals
n n
Sr ∑ ei∑2
(yi −)a 0 −1a
2

= i=1 =i=1 x i

xi yi − xi y2i
a1 = n
∑n∑ xi2 ∑∑∑ ( )
xi
−
a0 = y − a1x

Testing the goodness of fit

n

S t−
St ∑i (y − y 2

2
r = i=1
)
r = S n
St −ax 2

Sr =i=1 (y −
i 0 1

a
∑ ) i
 A MATLAB code for linear regression
function [a, r2] = linregr(x,y)
% input:
% x = independent variable
% y = dependent variable
% output:
% a = vector of slope, a(1), and intercept, a(2)
% r2 = coefficient of determination
n = length(x);
if length(y)~=n, error('x and y must be same length');
end x = x(:); y = y(:); % convert to column vectors
sx = sum(x); sy = sum(y);
sx2 = sum(x.*x); sxy = sum(x.*y); sy2 = sum(y.*y);
a(1) =
(n*sxy-sx*sy)/(n*sx2-sx^2); a(2)
= sy/n-a(1)*sx/n;
r2 = ((n*sxy-sx*sy)/sqrt(n*sx2-sx^2)/sqrt(n*sy2-sy^2))^2;
% create plot of data and best fit
line xp = linspace(min(x),max(x),2);
yp = a(1)*xp+a(2);
plot(x,y,'o',xp,yp
) grid on
CE 206: Engg. Computation sessional
 Fitting a straight line
1600
>> [a, r2] = linregr(v,F)
1400
a =
1200
19.4702 -234.2857
1000
r2 =
800
0.8805
600

F = −234.2857 +19.4702v 400

200

3.5
0

-200
10 20 30 40 50 60 70 80
3

2.5 Log-transformed data

>> [a, r2] =
2 linregr(log10(v),log10(F)
) a =
1.5 1.9842 -0.5620
r2 =
1
1.1 1.2 1.3 1.4 1.5 1.6 1.7 2
0.9481
1
1.8 1.9
Example:

Input :

>>rainfall = [10 15 20 25 30 35 40 45 50 55];

erosion = [0.8 1.1 1.4 2.0 1.8 2.5 2.3 3.0 2.7 3.3];

>>[a, r2] = linregr(rainfall, erosion)

Output:

>>a = 0.0525 0.3842

>> r2 = 0.9331
Variety of linear regressions
To fit a second order polynomial
n n
Sr ∑ ei2∑( ) yi − a 0 −1 ai x 2− 2 2

i=1 i=1 ax i
= =
Higher order polynomials
n n
Sr ∑ e∑( 2
i yi − a0 −1a x i− a 2x −L − am x
2 m 2
)
= i=1 = i=1 i
i

Multiple linear regression

y = a0 + a1x1 + a2 x2 +L am
xm n 2 n
Sr ∑ e ∑( i yi − a 0 −1 a1,ix 2 −2,i −L am
)
2
i=1 i=1 ax x m,i
= =
 Generalizing least squares
All the regression models you've seen (straight line, polynomials, multiple variables) can be
described using a unified mathematical framework based on basis functions and matrix
algebra.
 Generalizing least squares
Linear, polynomial, and multiple linear regression all
belong to the general linear least-squares model:
y = a0 z0 + a1z1 + a2 z2 +L amzm + e
where z0, z1, …, zm are a set of m+1 basis functions and
e is the error of the fit
y = Z a + e
{ } [ ]{ } { }
⎡z z11 L z ⎤
z ⎥
01 m1
z12 L
⎢ ⎥
m2
[Z ]= M M O
⎢z ⎢⎣z z M
02 0 z1n L mn

zji represents the then value of the⎥⎦jth basis function

calculated at the ith point.
Generalizing least squares
{y}= [Z]{a}+{e}
Applying the condition of least squares
⎛
⎞
n n m

∑ e∑ ⎜ yi − a∑z
2
2
Sr i j ⎟
= i=1 = i=1 ⎝ j=0 ⎠ ji
We get the following:
[[Z ] [Z ]]{a}= {[Z ] {y}}
T T

We can solve for the coefficients {a} using matrix

manipulations

To calculate residuals: {y} - [Z]{a}

 Example: general least squares
Fit the wind tunnel data to a 2nd order polynomial
v, m/s 10 20 30 40 50 60 70 80
F, N 25 70 380 550 610 1220 830 1450

Create the Z matrix:

>>v=[10, 20, 30,
40, 50,60,70, 80];
>>v=v(:);
>>Z=[ones(size(v)) v v.^2]
Z =
1 10 100
1 20 400
1 30 900
1 40 1600
1 50 2500
1 60 3600
1 70 4900
1 80 6400
Example: general least squares
Determine the coefficient matrix:
>>Z’*Z
ans =
8 360 20400
360 20400 1296000
20400 1296000 87720000

>>F=F(:);
>>Z’*F
ans =
5135
>>a=(Z’*Z)\(Z’*F)
312850 a =
20516500 -178.4821
16.1220
0.0372
polyfit and polyval function
MATLAB has a built-in function polyfit that fits a
least-squares nth order polynomial to data:
p = polyfit(x, y, n)
x: independent data
y: dependent data
n: order of polynomial to fit
p: coefficients of polynomial f(x)=p1xn+p2xn-1+…+pnx+pn+1

MATLAB’s polyval command can be used to

compute a value using the coefficients.
y = polyval(p, x)
Nonlinear regression
Not all fits are linear equations of coefficients and
basis functions.

Problems in transformation:
-Not all equations can be transformed easily or at all
-The best fit line represents the best fit for the transformed
variables, not the original variables.
Alternate approach: nonlinear regression.

Methodology:

-Write down the basis function.

-use MATLAB’s fminsearch function to find the
values of the coefficients where a minimum occurs
 Example: nonlinear regression
For the wind tunnel data, determine the coefficients
for the fit: F = a v a2 1

The m-file function: This function computes

function f = fSSR(a, xm, ym) the Sum of Squared
Residuals (SSR) between
yp = a(1)*xm.^a(2); the observed data and
f = sum((ym-yp).^2); model predictions:

Then, use fminsearch to obtain the values of a that

minimize fSSR:
a = fminsearch(@fSSR, [1, 1], [], v, F)
a = Initial guess for a1 and a2 Option Data
placeholder
2.5384 1.4359
Comparison between regressions
F = 0.2741v1.9842 transformed
a2
F = a 1v
F = 2.5384v1.4359 untransformed
2000

1500
untransformed
F (N)

1000

500
transformed
0
0 20 40 60 80
v (m/s)
 Polynomial interpolation
(1) Use MATLAB’s polyfit and polyval function
- the number of data points should be equal to the number of coefficients

(2) Newton’s interpolating polynomial

fn−1 x = b1 + b2 x − x1 +L + bn x − x1 x − x2 L x − xn−1
() ( ) ( )( ) ( )
b1 = f x1
( )
b2 = f x2 , x1 Divided
b3 = f [x3, x2 ,] x1 Differences
[ ]
M
bn = f xn , xn−1,L , x2 , x1
[
(3) Lagrange’s interpolating
] polynomial
Polynomial interpolation
Use polyfit and polyval function to predict the population
of 2000 using a 7th order polynomial fitted with the first 8 points
Year 1920 1930 1940 1950 1960 1970 1980 1990 2000
Populat 106.46 123.08 132.12 152.27 180.67 205.05 227.23 249.46 281.42
i on
(mil)

Also try using:

(a) Newton’s interpolating
polynomial
(b) Using a 1st order
polynomial using the
1980 and 1990 data
(c) Using a 2nd order
polynomial using the
1980,1990 and 1970
data
 Piecewise interpolation in MATLAB
The spline command performs cubic interpolation, generally
using not-a-knot method
yy=spline(x, y, xx)

Example: interpolate the Runge’s function f(x) = 1/(1+25x2)

taking 9 equidistant points between -1 and +1 using the spline
command.

x = linspace(-1, 1, 9);
y = 1./(1+25*x.^2);
xx = linspace(-1, 1);
yy = spline(x, y, xx);
yr = 1./(1+25*xx.^2)
plot(x, y, ‘o’, xx, yy, ‘-’, xx, yr, ‘--’)
MATLAB’s interp1 function

MATLAB’s interp1 function can perform several

different kinds of interpolation:

yi = interp1(x, y, xi, ‘method’)

x. and y contain the original data
xi. contains the points at which to interpolate
‘method’ is a string containing the desired method:
‘nearest’ - nearest neighbor interpolation ‘linear’
- connects the points with straight lines ‘spline’ -
not-a-knot cubic spline interpolation
‘pchip’ or ‘cubic’ - piecewise cubic Hermite interpolation
 Example: interp1 function
Time series of spot measurements of velocity:
t 0 20 40 56 68 80 84 96 104
v 0 20 20 38 80 80 100 100 125

140

120

100

>> 80
tt=linspace(0,110);
>> 60

v1=interp1(t,v,tt);
40
>>
plot(t,v,'o',tt,v1)
20

0
0 20 40 60 80 100 120