Programming with MATLAB

Clodomiro Ferreira EUI February 10, 2011 Aleksei Netsunajev

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

1 / 75

Preliminaries (I)

MATLAB knowledge we assume: very basic - basic operations and matrix manipulation, feeling confortable with the main MATLAB windows... Objective of the tutorial: Allowing you to feel confortable when solving computational and numerical problems in MATLAB

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

2 / 75

Preliminaries (II)

Levels of programming: Mathematica / Maple MATLAB /Gauss Fortran / C++

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

3 / 75

Tentative Outline
1

Before we begin M-les, Scripts and Functions Control of Flow: if, for, while, ... Programming: general issues Graphics in MATLAB MATLAB Help Applications
MATLAB Tutorial 2011 February 10, 2011 4 / 75

Ferreira, C. and A. Netsunajev ()

Working environment

Command Window Command History Workspace Current Directory

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

6 / 75

Working environment

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

7 / 75

Variables in MATLAB: Matrices or Arrays

(Most) objects you dene are understood as n-dimensional arrays by MATLAB You do need to: dene a name for the object. You dont need to: dene the dimensions of the object. Example: write in the command window a=3 MATLAB creates a 1x1 array named a. Check: write size(a) Tip: use the function size()!
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 9 / 75

Dening Matrices explicitly

More...

M = [1, 2; 3, 4; 5, 6] is a 3x2 matrix 1 2 M= 3 4 5 6 M = [low : step : high] creates a row vector with rst element low, last element high and distance between elements step M = linspace(0,1,5) creates a row vector with 5 elements M = (0, 0.25, 0.5, 0.75, 1) Remember: use the semi-colon ";" after a command in order to tell MATLAB not to show output on the Command window.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 10 / 75

Accessing sections of a Matrix

More...

1 2 M= 3 4 5 6 Accessing element (i, j): M (i, j). Type b = M (1, 2) b=2 Accessing row i: M (i, :) Accessing a particular range: M (i1 : i2 , j1 : j2 )

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

11 / 75

Useful Matrices and operations

More...

N1 N2 matrix of ones: ones(N1 , N2 ) N1 N2 matrix of zeros: zeros(N1 , N2 ) N1 N2 identity matrix : eye(N1 , N2 ) If M is an n m matrix, many common operations are available as MATLAB commands: inv(), det(), eig() Useful 1: MATLAB allows you to work on an "element by element" basis. Just add a "dot" n front of the operator: 1 4 M. M = 9 16 25 36 Useful 2: Multidimensional arrays. MATLAB allows you to create arrays with more than two dimensions. For example, A=zeros(2,2,3) creates a "cube" formed by three 2 2 arrays.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 12 / 75

M-les & Scripts: writing your own programs

Main tool for writing code in MATLAB. For simple problems, entering your requests at the Matlab prompt is fast and ecient. However, as the number of commands increases typing the commands over and over at the Matlab prompt becomes tedious. Similar advantages to a .do le in Stata. All built-in commands (i.e. mean(.), sqrt(.), inv(.), etc) are .m les. Two types of .m les:
Script les: do not take imput or retur/output arguments function les: may take imput / return arguments

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

14 / 75

M-les & Scripts: writing your own programs

In order to create and run an .m le, you need to:
FileNew M-file. File with a .m extension. Give it a name. Be sure the name is not an existing function!! >> help clodo clodo.m not found. Write your program / instructions.
Inside an .m le, you can "call" other .m les. Write comments on your program!

Save it on the current directory (cd). "call it": type on the Command Window clodo (or run clodo)

Variables and output created when running the .m le will be stored on the Workspace.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 15 / 75

Script File: simple example

This program generates pseudo-random sequence of 0 and 1 clc clear all L = 10 ; x = rand(1,L) ; y = round(x) ; z = sum(y,2) ; y z

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

17 / 75

Script File: simple example

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

18 / 75

Built-in & own Functions

Functions: .m les that can accept imput arguments and return output arguments.
Built-in: functions already existing in MATLAB. Example: inv(.), regress(.), plot(.), etc. Own functions: functions created by you

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

20 / 75

Built-in function: regress

Type help regress

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

21 / 75

Built-in function: regress

More useful functions... Two things to note: 1 Imput arguments (X , y ), can have dierent dimensions 2 A function can have one or more imput arguments (with a maximum) and one or more output arguments. 1 2

b=regress(y,X) [b,bint,r,rint,stats]=regress(y,X)

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

22 / 75

Own Functions
Useful when we want to automatize a particular set of operations which require imput arguments from another set of operations.

General structure:
function f = myfun(x,y, ...) commands ... f = expression; or with more output arguments function [f1,f2] = myfun(x,y, ...) commands ... f1 = expression; f2 = expression;
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 23 / 75

Own Functions: key features

All function .m les start with the command function f is the output. Can be replaced by [f,z,w,...], i.e. more than one outputs Unless you specify it explicitely (using local and global commands, variables (and their names) used within the function are are not stored in the workspace, and are just recognized inside this function. Key: You need to save the .m le with the name myfun After you specied a set of commands, you need to explicitely specify the output; hence the last line f=...

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

24 / 75

Own Functions: example 1

Lets create a function that evaluates the expression x +y 2 3 f (x, y ) = x + y + 2 function result = funct1 ( x , y ) imputs: result result = x^2 + y^3 + sqrt(x+y) / 2 ; x,y ; output:

Then, if we type in the Command Window (or we call it from another .m le) funct1(1,2) we get ans = 9.8660
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 25 / 75

Own Functions: example 1

More remarks: The name of the imputs, x and y, are only valid within the function. Usually, such imputs come fro previous calculations or parameters dened within a "main" program.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

26 / 75

Control of Flow

MATLAB has four basic decives a programmer can use to control de ow of commands:
for loops if-else-end constructions while loops switch-case constructions

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

28 / 75

Repeating with for loops

For loop repeats a group of statements a xed, predetermined number of times.

General structure:
for k=array ... end

Simple example
x=zeros(1,5) row of zeros to store for n=1:5 x(n)=n^2; end x= 1 4 9 16 25
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 30 / 75

You can nest for loops...

for loops can be nested: for i=1:4
for j=1:4
x(i,j)= i * j ;

end

end this will generate the following matrix: 1 2 x= 3 4

Ferreira, C. and A. Netsunajev ()

2 3 4 4 6 8 6 9 12 8 12 16
February 10, 2011 31 / 75

MATLAB Tutorial 2011

... but should avoid nesting whenever possible!!

MATLAB comparative advantage: vectorization and working with matrices. Nested loops are much slower than working with vectors the previous nested loop can be simplied: i=1:4; row vector 1 2 3 4 x=i * i ; vector multiplication. Careful with dimensions Also: always predene the matrix where you want to store

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

32 / 75

Repeating with while loops

This loop is used when the programmer does not know the number of repetitions a priori

General structure:
d = d0 Initialize variable d while expression with d ... d = ... update d end Useful for iterations on recursive probems...

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

34 / 75

Repeating with while loops: example

Simple xed point problem:
x0=0.5; initial value d=1 ; distance. Will be updated tol=0.0001 ; tolerance value while d>tol x1 = sqrt ( x0 ) ; d = abs ( x1-x0 ) ; update of distance x0 = x1 end x1=0.9999 d=8.4602e-005

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

35 / 75

Repeating with while loops: example

Figure: Fixed point

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

36 / 75

if-else-end constructions
Do some operations if some conditions hold.

General structure:
One alternative if expression ... end More than one alternative if expression 1 ... elseif expression 2 ... else ... end
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 38 / 75

Relational and logical operators

Operator

Description Relational >, < greater / lower than >=, <= greater / less or equal == equal ~= not equal Logical & and | or ~ not

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

39 / 75

Example: Simulating a Markov chain w/ 2 states

T=10; no. of periods P=[0.7,0.3;0.4,0.6]; transition matrix Y=[0.9,1.1]; possible values of state u=rand(1,T); draws from a U(0,1) z=zeros(1,T); t=1; initialization z(1)=Y(1); initialization index0=1; index for current state while t<T if u(t)<P(index0,1) z(t)=Y(1); index0=1; else z(t)=Y(2); index0=2; end t=t+1; end

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

40 / 75

Theswitch-case construction
Switch compares the input expression to each case value. Once the match is found it executes the associated commands. For most practical cases, it achieves similar results to if-elseif-else constructions

General structure:
switch expression scalar or string case value1 executes if expression = value1 commands... case value2 commands... ... otherwise commands end

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

42 / 75

Control of ow: programming tips

Try to program inside out: start with the inner section of your code, check it produces the desired results, and proceed towards the outer loops. In this way, you keep track of each step and possible errors. Ask MATLAB to print intermediate results / variable values. This is a good way to know exactly what is going on inside your code. Check the Workspace window: sometimes, either the dimensions of your matrices are not the ones you though... or matrices are just empty!!!

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

43 / 75

(Some) "common" error messages

???Error using ==> minus Matrix dimensions must agree. : Often it is an indexing mistake that causes the terms to be of dierent size. ??? Error using ==> mtimes Inner matrix dimensions must agree.: Note the dierence between this error and the previous one. This error often occurs because of indexing issues OR because you meant to use componentwise multiplication but forgot the dot. Error: Unbalanced or misused parentheses or brackets. : for a complex expression you have to go through it very carefully to nd your typo. ??? Error using ==> Too many input arguments. : Input arguments must be in a format expected by the function. This will be very function-specic.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 44 / 75

Logical Addressing in MATLAB

You can solve some tricky problems using some logical addressing. Two useful functions / operations:
find(.) nds indices of non-zero elements in an array (expression) acts as an indicator function: it takes value = 1 if expression holds

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

45 / 75

Logical Addressing: example

1 1 3 A= 5 9 2 4 4 6
1

x =rand(1,3) = 0.1576 0.9706 0.9572

Find the indices where A>=4: ind1 = find(A >= 4) ind1 = 2 3 5 6 9 Operate over values of x that satisfy certain condition: z = (x.^2) . (x > 0.2) z = (0, 0.9421, 0.9162)

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

46 / 75

Basic commands
Be careful with function sqrt() using matrices. Consider example: 2 2 A= ; 2 2 B = sqrt(A) 1.4142 1.4142 B= ; 1.4142 1.4142 4 4 B B = 4 4 C = chol (A); 2 2 C C = 2 2

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

48 / 75

Programming

A variable is a tag that you assign to a value while that value remains in memory. You refer to the value using the the tag. You do not need to type or declare variables. MATLAB variable names must begin with a letter. MATLAB is case sensitive, so A and a are two dierent variables! Do not name a variable using a reserved names, such as i, j, mode, char , size and path.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

49 / 75

Programming. Structures

Structures are multidimensional MATLAB arrays. This is very much like a database entity. Structures are useful to group variables. Let structure consist of 3 variables: Student.name, Student.score, Student.grade. The whole structure could be an input for user-dene function. It is convenient to use the structures when you have a lots of variables and you use your own functions. You wont need to pass all 3 variables to the function, but just the whole structure See provided example.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

51 / 75

Programming. Local variables

Each MATLAB function has its own variables. These are separate from those of other functions, and from those of the base workspace, hence they are called local.They live only while the function is running. Scripts do not have a separate workspace. They store variables in a workspace that is shared with the caller of the script. When called from the command line, they share the base workspace. When called from a function, they share that functions workspace. If you run a script that alters a variable that already exists in the workspace, that variable is overwritten by the script.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

53 / 75

Programming. Global variables

If several functions, and the base workspace, all declare a particular name as a global variable, then they all share a single copy of that variable. Any assignment to that variable, in any function, is available to all the other functions declaring it as global. Instead of using a global variable, you may want to pass the variable to other functions as an additional argument. In this way, you make sure that any shared access to the variable is intentional. If you have to pass a number of additional variables, you can conveniently put them into a structure and pass it as one argument. See provided example.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 54 / 75

Programming. Debugging

Do you think your program is not producing the results that you expected? Then you can debug your program and see whats wrong. The standard debug tool are the breakpoints. Set breakpoints to pause execution of your program so you can examine values of variables where you think the problem can be. After setting breakpoints, run the le

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

56 / 75

Programming. Debugging
Then the Debug menu allows to: Run Commence execution of le and run until completion or until a breakpoint is met. Go Until Cursor Continue execution of le until the line where the cursor is positioned. Also available on the context menu. Step Execute the current line of the le. Step In Execute the current line of the le and, if the line is a call to another function, step into that function. Continue Resume execution of le until completion or until another breakpoint is encountered. Step Out After stepping in, run the rest of the called function or subfunction, leave the called function, and pause. Exit Debug Mode Exits debug mode.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 57 / 75

2D Plots
There are many tools and ways for creating and editing your plots, both from the command line and by using the menus of Matlab (interface in the Figure window). It is possible to export your graph to nearly all conventional formats (.pcx, .bmp, .jpg, .pdf) via Save As option. See provided example on plots and subplots.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

59 / 75

3D Plots
The primary method to create the 3D plot is the surf command which is used in combination with the meshgrid command. Meshgrid creates a matrix of (x , y) points over which the surface is to be plotted. See provided example on 3D plots.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

61 / 75

MATLAB help
Matlab has the very user-friendly and extensive built-in and on-line help system. To access built-in help Type help in the Command Window Press F1 Go to Help menu Online users guide is available at http://www.mathworks.com/help/techdoc/matlab_product_ page.html Google the function you need end exploiting the web resourses available

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

62 / 75

Application 1: Bisection Algorithm

We have a consumption saving problem with idiosyncratic uncertainty (but no aggregate uncertainty), borrowing constraints and no labor decision. The standard euler equation is given by u (ct ) = (1 + r )Et u (ct+1 ) (1) and the period budget constraint ct = st wt + (1 + r )kt kt+1 Here, given no labor decisions and no aggregate uncertainty, w (labor wage) and r (net real interest rate) are time invariant. st is the current labor productivity state. (1) is usually non-linear. We wish to nd the value of kt+1 that, for given kt, st solves the Euler equation. To do this, one strategy is using a bisection algorithm
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 64 / 75

Optimization toolbox: FMINCON

FMINCON is the function with the most features hence we base the example on it. Other optimization functions are less complex and work in a similar way. The function is designed to nd minimum of constrained nonlinear multivariate function: c(x) 0 ceq(x) = 0 Ax b minx f (x)s.t. Aeq x = beq lb x ub where x, b, beq, lb, ub are vectors, A and Aeq are matrices, c(x) and ceq(x) are functions that return vectors, and f (x) is a function that returns a scalar.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 66 / 75

Optimization toolbox: FMINCON. Examples

Example 1. Find values of x that minimize f (x) = x1 x2 x3 , starting at the point x = [10; 10; 10], subject to the constraints: 0 x1 + 2x2 + 2x3 72. Example 2. Minimize log-likelihood function: l (B, 2 , ..., , M ) = T log det(B) +
M 1 2 T

B 1 B 1
T

1t|T ut ut
t=1

+
m=2

Tm 2

log det(m ) + 1 tr 2

B 1 1 B 1 m

mt|T ut ut
t=1

with respect to matrices m and B, taking other parameter as given, 0 0 0 possibly subject to B = and all elements of diag (m ) are 0.01.
Ferreira, C. and A. Netsunajev () MATLAB Tutorial 2011 February 10, 2011 67 / 75

Optimization toolbox: FSOLVE

The function fsolve is meant to solve system of nonlinear equations Syntax: [X , fval ] = fsolve(fun, x0, options) fsolve nds a root (zero) of a system of nonlinear equations. Example: Find a matrix x that satises the equation xxx = at the point strt0 = 1 1 1 1 . 1 2 3 4 starting

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

68 / 75

Dening matrices

Return

Output x=123 1 x = [1;2;3] x= 2 3 1 2 3 A = [1 2 3; 4 5 6] A = 4 5 6

Input x = [1 2 3]

Comments row vector column vector 2 x 3 matrix

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

69 / 75

Accessing matrices
Return

Input [1 2 3; A = 4 5 6; 7 8 9]; A(2,3) A(:,3) A(2,:) A(1:2,2:3)

Output supressed ans = 6 3 ans = 6 9 ans = 4 5 6 2 3 ans = 5 6

Comments create matrix element in 2nd row, 3rd col 3rd col 2nd row block

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

70 / 75

Special matrices
Return

Input x = zeros(2,4) x = ones(2,3) A = eye(3)

Output 0 0 0 0 x= 0 0 0 0 1 1 1 x= 1 1 1 1 0 0 A= 0 1 0 0 0 1

Comments 2x4 matrix of zeros 2x3 matrix of ones 3x3 identity matrix

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

71 / 75

Useful Built-in functions (I)

error(error message): displays error message and abort function when a certain condition is satised tic: starts a time counter. t = tic assigns the current time to the variable t toc: this will display time elapsed since tic was called. fprint(abc): prints text abc on the Command Window size(.) : rturns matrix / array dimensions rand(n,m): generates an n m matrix of pseudo-random numbers from a U[0, 1] sort(X,dim): sorts elements of array X along dimension dim

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

72 / 75

Useful Built-in functions (II)

floor(x): rounds x towards minus innity ceil(x): rounds x towards plus innity round(x): rounds x towards nearest integer Let x = [0.2234 , -1.4434 , 5.3789]. Then: floor(x) = [0 , -2 , 5]. ceil(x) = [1 , -1 , 6]. round(x) = [0 , -1 , 5].

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

73 / 75

(Pseudo-) Random Numbers in MATLAB (I)

Two built-in functions to generate pseudo-random numbers: 1 rand(.); uniformly [0,1] distirbuted pseudo rn.
1 2 2 3

a = rand; generates a scalar-random number A = rand(n,m) generates an nxm matrix of random numbers

randn(.); (standard) normally distributed pseudo-rn rand(state,0) or randn(state,0) useful to repeat a computation using the same sequence of pseudo- random numbers.

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

74 / 75

(Pseudo-) Random Numbers in MATLAB (II)

Return

Two alternative ways of generating normally distributed (pseudo) rn: 1 probability integral transform property: if U is distributed uniformly on (0,1), then 1 (U) will have the standard normal distribution. 2 (approximation) Central Limit Theorem! Generate 12 U [0, 1] , add them up, and substract 6. (11th order polynomial approx. to the normal distribution)

Ferreira, C. and A. Netsunajev ()

MATLAB Tutorial 2011

February 10, 2011

75 / 75