An Introduction to
Matlab
For CSE 802: Pattern Recognition
Feng Kang
[email protected]
Start Matlab
You
can access it from CSE lab but its more
easy to go to engineering lab and use
Matlab there.
Machines
in engineering buildings labs.
Start->All Programs->Matlab 7.0.
License
issues, especially for some toolbox.
Exit Matlab if you do not use it.
Topics
Data
structure of Matlab.
Some
useful Matlab functions for this course.
Plotting
Two
of data.
examples:
Plotting of multivariate Gaussian data.
PCA: compute PCA and plot the data of reduced
dimensionality.
Scalars, Vectors and Matrices
Scalar:
Just a number: a = 1; b = 3;
Vector:
Column vector: a = [1; 2; 3; 4].
Row vector: b = [1 2 3 4].
Transpose: a = b;
Matrix:
A = [1 2 3; 4 5 6; 7 8 9];
Access Elements in Matrices
Access
a single element.
A[row index, colum index]
A[1,3] = 3;
Access
a sub-matrix.
Extract out part of rows: B = A[1:2, :];
Extract out part of columns: C=A[:,1:2];
Operations on Matrix
Cell
by cell operation.
.
E.g. B = A.^2;
B= [1 4 9; 16 25 36; 49; 64; 81];
Matrix
operation.
+, -, *.
Control Structures for Matlab(1)
Conditional
if expression1
statements1
elseif expression2
statements2
else
statements3
end
Example
if (a>3)
b=4;
end;
statements.
Control Structure for Matlab(2)
Loop
structure: for loop
for variable = expression
statements
end
j=0;
for i=1:10
j = j+i;
end
Loop
structure: while loop.
while expression
statements
end
Symbolic Toolbox(1)
Declare
a symbol object.
Not a number but a symbol.
Syntax: syms arg1 arg2 real.
Use symbols to represent a function.
syms x u real
syms s positive
f = exp(-(x-u)^2/s^2);
Symbolic Toolbox(2)
Manipulate
the function.
Compute integration.
g = int(f,x, -inf, inf); result: g =s*pi^(1/2)
Gaussian distribution: f/g
There
are many other ways to manipulate
the functions: e.g. differentiation.
Load and save data
Load
data:
Matrix format: load(file path);
Save
data:
Matrix format: save(file path, matrix name, ascii);
Common Functions in CSE 802
Functions related to Multivariate Gaussian distribution.
mean(A)
cov(x), cov(x,y); x, y are vectors.
inv(A): inverse of the matrix.
det(A): determinant of the matrix.
mvnrnd(mu, sigma, num of data.)
Functions related to dimensionality reduction.
eigs(A): compute eigenvector of A.
Plotting
Plot
function:
Plot one line: plot(X1,Y1,LineSpec).
Plot several lines on the same figure:
figure(1);
hold
on;
plot(x1, y1, LineSpec1);
plot(x2, y2, LineSpec2);
hold off;
legend(line 1, line 2, );
xlabel(description of x axis); ylabel(description of y axis);
Plotting Example
x = 1:10 ;
y = 3*x;
z = x.^2;
figure(1)
hold on;
plot(x, y, '-ro');
plot(x, z, '-b*');
hold off;
legend('y=3*x', 'z=x.^2');
xlabel('x');
ylabel('function values');
Ezplot
Mainly
used for implicitly defined functions.
Sometimes, its more convenient to plot the implicit
form of the functions.
E.g.
Function format: ezplot(f,[xmin,xmax,ymin,ymax]) plots
f(x,y) = 0 over xmin < x < xmax and ymin < y < ymax.
The
first parameter f is passed as a string.
Ezplot Example
figure(1)
ezplot('x^2+y^2-1',[-1,1,-1,1]);
xlabel('x');
ylabel('function values');
Generate Multivariate Gaussian Data
Generate
multivariate Gaussian data.
rand_data = mvnrnd(mu, sigma, num of data.)
E.g.
mu1 = [0 0];
sigma1 = [1 0; 0 1];
r1 = mvnrnd(mu1, sigma1, 50);
plot(r1(:,1), r1(:,2), *);
PCA to Extract the Major Information
in Data and Plot it.
PCA
to reduce dimensionality of data and
plot them in 2-D space.
Example:
IRIS data:
four dimensional data. Hard to visualize.
Apply PCA to reduce to two dimensional data
and plot them.
Example codes of PCA
Codes:
X = load('iris_data');
c = mean(X);
X = X - repmat(c, size(X,1), 1);
covar = cov(X);
opt.disp = 0;
[p, D] = eigs(covar, 2, 'LA', opt);
reduced = X*p;
figure(1)
hold on;
plot(reduced(1:50, 1), reduced(1:50, 2), 'o');
plot(reduced(51:100, 1), reduced(51:100, 2), '*');
plot(reduced(101:150, 1), reduced(101:150, 2), '+');
hold off;
legend('Setosa', 'Versicolour','Virginica');
Plot of PCA
