Computational Techniques

Module 5: Regression and Interpolation

Dr. Niket Kaisare

Department of Chemical Engineering
Indian Institute of Technology - Madras
 Regression Example
 Given the following data:
x 0.8 1.4 2.7 3.8 4.8 4.9
y 0.69 1.00 2.02 2.39 2.34 2.83

Regression:
Ob
Obtain a straight
h line
l
that best fits the data
 Interpolation Example
 Given the following data:
x 0.8 1.4 2.7 3.8 4.8 4.9
y 0.69 1.00 2.02 2.39 2.34 2.83

Interpolation:
“J
“Join the
h dots”
d ” and
d
find a curve passing
through the data.
Regression vs. Interpolation
x 0.8 1.4 2.7 3.8 4.8 4.9
y 0.69 1.00 2.02 2.39 2.34 2.83

 In regression,
regression we are interested in fitting a
chosen function to data
y = 0.45
0 45 + 0.47x
0 47x

 In interpolation,
p , ggiven finite amount of data,, we
are interested in obtaining new data-points
within this range.
At x = 2.0, y = 1.87
 Example: Kinetic Rate Constants
(Regression)
R i )
 Experiments with conversion measured at
various temperatures
6
E
r  k0e ( RT )
ca 4

log(rr)
0 Slope = E/R

E 1
ln(r )  ln(k0 )     ln(ca )
-2

R T  -4
Intercept = ln(k0)
-3.5 -3 -2.5 -2 -1.5
-1/T -3
x 10
Example: Viscosity of Oil
(Interpolation)
I l i )
 Viscosity of lubricant oil was measured between
-20 to 200 degrees Celcius in steps of 20 oC.
 Interpolation is used if viscosity is desired at an
intermediate temperature
General Setup
 Let x be an independent variable
and y be a dependent variable

 Given the data:

 x1, y1 ,  x2 , y2 , ,  x N , y N 
Find parameters θ to get a “best‐fit” curve
y  f  x; 
Regression vs. Interpolation
Regression Interpolation
 Choose a function  Various standard
form for f(x;θ) function forms exist
 For a given θ, obtain  The interpolating
the values ŷˆ i from function passes
the model through all the points
 The best θ minimizes  Can be used to “fill-
the error yi  yˆi in” the data at new
points
Regression vs. Interpolation
 “Curve Fitting”
g  “Joining
J g the dots”

 Obtain a functional  Obtain the value of y

form to fit the data at intermediate point
p
Outline: Regression
 Linear Regression in One Variable

 Linear Regression in Multiple Variables

 Polynomial Regression

 A l i and
Analysis d EExtension
i

 Non-Linear Regression
g
Linear Regression: One Variable
 Model: y  f  x; 

 Actual Data:  x1, y1 ,  x2 , y2 , ,  x N , y N 

 P d
Prediction: yˆi  f  xi ; 

 Errors: ei  yi  yˆi

yi  f  xi ;   ei

 
yˆ i

 Mean / Variance:
 i
 2
x  xi sx 
x  x
( N  1)   sx
N
Extension to Multi
Multi--Variables
 Let x1, x2, …, xn be n variables.
Let there be N data points for each:
 x11, x21,, xn1; y1 ,
 x12 , x22 ,, xn 2 ; y2 ,

 x1N , x2 N ,, xnN ; y N 

Obtain θ for

y  f x; θ 
Linear Regression (multi
(multi--variable)
 Data  xi , ui , wi ; yi 
 Model y  a0  a1x  a2u  a3w

 Error ei  yi  a0  a1xi  a2ui  a3wi 

 Least Squares Criterion

2
N  

min  yi  a0  a1xi  a2ui  a3wi 
a0 , a1 , a2 , a3 i 1    
 ei 
Linear Regression (alternate)
1 x1 u1 w1   a0   y1 
1 x u w  a   y 
 2 2 2  1   2 

      a2    
1 x  a   y 
 u
NN
w N  3  N
X  Y

Least Squares

 X X T
1 T
X Y
Polynomial / Functional Regression
 Example:
p Specific
p heat as a function of T
◦ Methane: cp = 85.8 + 1.126e-2 T – 2.1141e-6 T2

 Example: Antoine’s vapor pressure relationship

b
◦ ln psat   a 
T c
 Polynomial / Functional Regression
1 x1 x12 x12 
 3
Model:  1 x2 x22 x2 
X
y  a0  a1x  a2 x 2  a3 x 3     
 2 3 
1 x N xN x N 

1 ln x1  1
x1 x1 
 
 1 ln x2  1 x2 
M d l
Model: X  x2
a2     
y  a0  a1 ln( x)   a3 x  
x 1 ln x N  1
xN xN 

Outline: Interpolation
 Polynomial fitting and limitations

 Lagrange interpolating polynomials

 Newton’s methods

 S li interpolation
Spline i l i
Lagrange Polynomials
Pi 
 x  x1  x  x2 ... x  xi 1  x  xi 1 ... x  x N 
xi  x1 xi  x2 ...xi  xi 1  xi  xi 1 ...xi  xN 
 x xj 
  
 
j  i  xi  x j 

The interpolating
polynomial becomes

f ( x)  y1P1  y2 P2  ...  y N PN
Newton’s
Newton s Divided Differences
yi 11  yi
yi  1, i  
xi 1  xi
y[i  2, i  1]  y[i  1, i ]
yi  2, i  1, i  
xi 2  xi
y[i  3, i  2, i  1]  y[i  2, i  1, i ]
yi  3, i  2, i  1, i  
xi 3  xi


c0  y1, c1  y2,1, c2  y3,2,1, c3  y4,3,2,1, ...

f ( x)  c0  c1 ( x  x1 )  c2 ( x  x1 )( x  x2 )  ...