**Przeglądaj wersję html pliku:**

### MATLAB array manipulation tips and tricks (ang)

M ATLAB array manipulation tips and tricks

Peter J. Acklam E-mail: pjacklam@online.no URL: http://home.online.no/~pjacklam 18th October 2003

Copyright © 2000–2003 Peter J. Acklam. All rights reserved. Any material in this document may be reproduced or duplicated for personal or educational use. M ATLAB is a trademark of The MathWorks, Inc. TEX is a trademark of the American Mathematical Society. Adobe, Acrobat, Acrobat Reader, and PostScript are trademarks of Adobe Systems Incorporated.

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 High-level vs low-level code 1.1 Introduction . . . . . . . . . . 1.2 Advantages and disadvantages 1.2.1 Portability . . . . . . . 1.2.2 Verbosity . . . . . . . 1.2.3 Speed . . . . . . . . . 1.2.4 Obscurity . . . . . . . 1.2.5 Difﬁculty . . . . . . . 1.3 Words of warning . . . . . . . v 1 1 1 1 2 2 2 2 2 3 3 4 5 6 6 6 6 7 7 7 7 8 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 10 10 11 11 11

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

2 Operators, functions and special characters 2.1 Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Built-in functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 M-ﬁle functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Basic array properties 3.1 Size . . . . . . . . . . . . . . . . . . . 3.1.1 Size along a speciﬁc dimension 3.1.2 Size along multiple dimensions 3.2 Dimensions . . . . . . . . . . . . . . . 3.2.1 Number of dimensions . . . . . 3.2.2 Singleton dimensions . . . . . . 3.3 Number of elements . . . . . . . . . . . 3.3.1 Empty arrays . . . . . . . . . . 4 Array indices and subscripts 5 Creating basic vectors, matrices and arrays 5.1 Creating a constant array . . . . . . . . . . . . . . . . . . . . 5.1.1 When the class is determined by the scalar to replicate 5.1.2 When the class is stored in a string variable . . . . . . 5.2 Special vectors . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Uniformly spaced elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6 Shifting 12 6.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 6.2 Matrices and arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

ii

CONTENTS

7 Replicating elements and arrays 7.1 Creating a constant array . . . . . . . . . . . . . . . . . . 7.2 Replicating elements in vectors . . . . . . . . . . . . . . . 7.2.1 Replicate each element a constant number of times 7.2.2 Replicate each element a variable number of times 7.3 Using KRON for replicating elements . . . . . . . . . . . 7.3.1 KRON with an matrix of ones . . . . . . . . . . . 7.3.2 KRON with an identity matrix . . . . . . . . . . . 8 Reshaping arrays 8.1 Subdividing 2D matrix . . . . . . . . . . . . . . . . . . 8.1.1 Create 4D array . . . . . . . . . . . . . . . . . . 8.1.2 Create 3D array (columns ﬁrst) . . . . . . . . . . 8.1.3 Create 3D array (rows ﬁrst) . . . . . . . . . . . 8.1.4 Create 2D matrix (columns ﬁrst, column output) 8.1.5 Create 2D matrix (columns ﬁrst, row output) . . 8.1.6 Create 2D matrix (rows ﬁrst, column output) . . 8.1.7 Create 2D matrix (rows ﬁrst, row output) . . . . 8.2 Stacking and unstacking pages . . . . . . . . . . . . . . . . . . . . . . .

iii 13 13 13 13 13 14 14 14 16 16 16 16 17 17 18 18 19 19 20 20 20 21 22 22 23 24 25 26 27 28 28 28 29 30 30 30 30 30 31 32 32 32 33 33 33 33 34

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

9 Rotating matrices and arrays 9.1 Rotating 2D matrices . . . . . . . . . . . . . . . . . . . . . 9.2 Rotating ND arrays . . . . . . . . . . . . . . . . . . . . . . 9.3 Rotating ND arrays around an arbitrary axis . . . . . . . . . 9.4 Block-rotating 2D matrices . . . . . . . . . . . . . . . . . . 9.4.1 “Inner” vs “outer” block rotation . . . . . . . . . . . 9.4.2 “Inner” block rotation 90 degrees counterclockwise . 9.4.3 “Inner” block rotation 180 degrees . . . . . . . . . . 9.4.4 “Inner” block rotation 90 degrees clockwise . . . . . 9.4.5 “Outer” block rotation 90 degrees counterclockwise 9.4.6 “Outer” block rotation 180 degrees . . . . . . . . . 9.4.7 “Outer” block rotation 90 degrees clockwise . . . . 9.5 Blocktransposing a 2D matrix . . . . . . . . . . . . . . . . 9.5.1 “Inner” blocktransposing . . . . . . . . . . . . . . . 9.5.2 “Outer” blocktransposing . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

10 Basic arithmetic operations 10.1 Multiply arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1.1 Multiply each 2D slice with the same matrix (element-by-element) . 10.1.2 Multiply each 2D slice with the same matrix (left) . . . . . . . . . 10.1.3 Multiply each 2D slice with the same matrix (right) . . . . . . . . . 10.1.4 Multiply matrix with every element of a vector . . . . . . . . . . . 10.1.5 Multiply each 2D slice with corresponding element of a vector . . . 10.1.6 Outer product of all rows in a matrix . . . . . . . . . . . . . . . . . 10.1.7 Keeping only diagonal elements of multiplication . . . . . . . . . . 10.1.8 Products involving the Kronecker product . . . . . . . . . . . . . . 10.2 Divide arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2.1 Divide each 2D slice with the same matrix (element-by-element) . . 10.2.2 Divide each 2D slice with the same matrix (left) . . . . . . . . . . . 10.2.3 Divide each 2D slice with the same matrix (right) . . . . . . . . . .

CONTENTS

11 More complicated arithmetic operations 11.1 Calculating distances . . . . . . . . . . . . . . . . . 11.1.1 Euclidean distance . . . . . . . . . . . . . . 11.1.2 Distance between two points . . . . . . . . . 11.1.3 Euclidean distance vector . . . . . . . . . . 11.1.4 Euclidean distance matrix . . . . . . . . . . 11.1.5 Special case when both matrices are identical 11.1.6 Mahalanobis distance . . . . . . . . . . . . . 12 Statistics, probability and combinatorics 12.1 Discrete uniform sampling with replacement . . 12.2 Discrete weighted sampling with replacement . 12.3 Discrete uniform sampling without replacement 12.4 Combinations . . . . . . . . . . . . . . . . . . 12.4.1 Counting combinations . . . . . . . . . 12.4.2 Generating combinations . . . . . . . . 12.5 Permutations . . . . . . . . . . . . . . . . . . 12.5.1 Counting permutations . . . . . . . . . 12.5.2 Generating permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

iv 35 35 35 35 35 35 36 36 38 38 38 38 39 39 39 39 39 40 41 41 41 42 42 42 42 43 43 43

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

13 Identifying types of arrays 13.1 Numeric array . . . . . . . . . . . . . . . . . . . 13.2 Real array . . . . . . . . . . . . . . . . . . . . . 13.3 Identify real or purely imaginary elements . . . . 13.4 Array of negative, non-negative or positive values 13.5 Array of integers . . . . . . . . . . . . . . . . . 13.6 Scalar . . . . . . . . . . . . . . . . . . . . . . . 13.7 Vector . . . . . . . . . . . . . . . . . . . . . . . 13.8 Matrix . . . . . . . . . . . . . . . . . . . . . . . 13.9 Array slice . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

14 Logical operators and comparisons 44 14.1 List of logical operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 14.2 Rules for logical operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 14.3 Quick tests before slow ones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 15 Miscellaneous 15.1 Accessing elements on the diagonal . . . . . 15.2 Creating index vector from index limits . . . 15.3 Matrix with different incremental runs . . . . 15.4 Finding indices . . . . . . . . . . . . . . . . 15.4.1 First non-zero element in each column 15.4.2 First non-zero element in each row . . 15.4.3 Last non-zero element in each row . . 15.5 Run-length encoding and decoding . . . . . . 15.5.1 Run-length encoding . . . . . . . . . 15.5.2 Run-length decoding . . . . . . . . . 15.6 Counting bits . . . . . . . . . . . . . . . . . Glossary . . . . . . . . . . . . . . . . . . . . . . . Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 46 47 48 48 48 49 50 50 50 51 51 52 53

CONTENTS

v

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 A M ATLAB resources 55

Preface

The essence

This document is intended to be a compilation of tips and tricks mainly related to efﬁcient ways of manipulating arrays in M ATLAB. Here, “manipulating arrays” includes replicating and rotating arrays or parts of arrays, inserting, extracting, replacing, permuting and shifting arrays or parts of arrays, generating combinations and permutations of elements, run-length encoding and decoding, arithmetic operations like multiplying and dividing arrays, calculating distance matrices and more. A few other issues related to writing fast M ATLAB code are also covered. I want to thank Ken Doniger, Dr. Denis Gilbert for their contributions, suggestions, and corrections.

Why I wrote this

Since the early 1990’s I have been following the discussions in the main M ATLAB newsgroup on Usenet, comp.soft-sys.matlab. I realized that many of the postings in the group were about how to manipulate arrays efﬁciently, which was something I had a great interest in. Since many of the same questions appeared again and again, I decided to start collecting what I thought were the most interestings problems and solutions and see if I could compile them into one document. That was the beginning of what you are now reading.

Intended audience

This document is mainly intended for those of you who already know the basics of M ATLAB and would like to dig further into the material regarding manipulating arrays efﬁciently.

How to read this

This document is more of a reference than a tutorial. Although it might be read from beginning to end, the best way to use it is probably to get familiar with what is covered and then look it up here when you bump into a problem which is covered here. The language is rather technical although many of the terms used are explained. The index at the back should be an aid in ﬁnding the explanation for a term unfamiliar to you.

vi

CONTENTS

vii

Organization

Instead of just providing a compilation of questions and answers, I have organized the material into sections and attempted to give general answers, where possible. That way, a solution for a particular problem doesn’t just answer that one problem, but rather, that problem and all similar problems. Many of the sections start off with a general description of what the section is about and what kind of problems that are solved there. Following that are implementations which may be used to solve the given problem.

Typographical convensions

All M ATLAB code is set in a monospaced font, like this, and the rest is set in a proportional font. En ellipsis (...) is sometimes used to indicated omitted code. It should be apparent from the context whether the ellipsis is used to indicate omitted code or if the ellipsis is the line continuation symbol used in M ATLAB. M ATLAB functions are, like other M ATLAB code, set in a proportional font, but in addition, the text is hyperlinked to the documentation pages at The MathWorks’ web site. Thus, depending on the PDF document reader, clicking the function name will open a web browser window showing the appropriate documentation page.

Credits

To the extent possible, I have given credit to what I believe is the author of a particular solution. In many cases there is no single author, since several people have been tweaking and trimming each other’s solutions. If I have given credit to the wrong person, please let me know. In particular, I do not claim to be the sole author of a solution even when there is no other name mentioned.

Errors and feedback

If you ﬁnd errors, or have suggestions for improvements, or if there is anything you think should be here but is not, please mail me and I will see what I can do. My address is on the front page of this document.

Chapter 1

High-level vs low-level code

1.1 Introduction

Like other computer languages, M ATLAB provides operators and functions for creating and manipulating arrays. Arrays may be manipulated one element at a time, like one does in low-level languages. Since M ATLAB is a high-level programming language it also provides high-level operators and functions for manipulating arrays. Any task which can be done in M ATLAB with high-level constructs may also be done with lowlevel constructs. Here is an example of a low-level way of squaring the elements of a vector

x = [ 1 2 3 4 5 ]; y = zeros(size(x)); for i = 1 : numel(x) y(i) = x(i)^2; end % % % % % vector of values to square initialize new vector for each index square the value end of loop

and here is the high-level, or “vectorized”, way of doing the same

x = [ 1 2 3 4 5 ]; y = x.^2; % vector of values to square % square all the values

The use of the higher-level operator makes the code more compact and more easy to read, but this is not always the case. Before you start using high-level functions extensively, you ought to consider the advantages and disadvantages.

1.2 Advantages and disadvantages

It is not always easy to decide when to use low-level functions and when to use high-level functions. There are advantages and disadvantages with both. Before you decide what to use, consider the following advantages and disadvantages with low-level and high-level code.

1.2.1 Portability

Low-level code looks much the same in most programming languages. Thus, someone who is used to writing low-level code in some other language will quite easily be able to do the same in M ATLAB. And vice versa, low-level M ATLAB code is more easily ported to other languages than high-level M ATLAB code. 1

CHAPTER 1. HIGH-LEVEL VS LOW-LEVEL CODE

2

1.2.2 Verbosity

The whole purpose of a high-level function is to do more than the low-level equivalent. Thus, using high-level functions results in more compact code. Compact code requires less coding, and generally, the less you have to write the less likely it is that you make a mistake. Also, is is more easy to get an overview of compact code; having to wade through vast amounts of code makes it more easy to lose the big picture.

1.2.3 Speed

Traditionally, low-level M ATLAB code ran more slowly than high-level code, so among M ATLAB users there has always been a great desire to speed up execution by replacing low-level code with high-level code. This is clearly seen in the M ATLAB newsgroup on Usenet, comp.soft-sys.matlab, where many postings are about how to “translate” a low-level construction into the high-level equivalent. In M ATLAB 6.5 an accelerator was introduced. The accelerator makes low-level code run much faster. At this time, not all code will be accelerated, but the accelerator is still under development and it is likely that more code will be accelerated in future releases of M ATLAB. The M ATLAB documentation contains speciﬁc information about what code is accelerated.

1.2.4 Obscurity

High-level code is more compact than low-level code, but sometimes the code is so compact that is it has become quite obscure. Although it might impress someone that a lot can be done with a minimum code, it is a nightmare to maintain undocumented high-level code. You should always document your code, and this is even more important if your extensive use of high-level code makes the code obscure.

1.2.5 Difﬁculty

Writing efﬁcient high-level code requires a different way of thinking than writing low-level code. It requires a higher level of abstraction which some people ﬁnd difﬁcult to master. As with everything else in life, if you want to be good at it, you must practice.

1.3 Words of warning

Don’t waste your time. Don’t rewrite code which doesn’t need rewriting. Don’t optimize code before you are certain that the code is a bottleneck. Even if the code is a bottleneck: Don’t spend two hours reducing the execution time of your program by one minute unless you are going to run the program so many times that you will save the two hours you spent optimizing it in the ﬁrst place.

Chapter 2

Operators, functions and special characters

Clearly, it is important to know the language you intend to use. The language is described in the manuals so I won’t repeat what they say, but I encourage you to type help help help help ops relop arith slash

at the command prompt and take a look at the list of operators, functions and special characters, and look at the associated help pages. When manipulating arrays in M ATLAB there are some operators and functions that are particularely useful.

2.1 Operators

In addition to the arithmetic operators, M ATLAB provides a couple of other useful operators : .’ ’ The colon operator. Type help colon for more information. Non-conjugate transpose. Type help transpose for more information. Complex conjugate transpose. Type help ctranspose for more information.

3

CHAPTER 2. OPERATORS, FUNCTIONS AND SPECIAL CHARACTERS

4

2.2 Built-in functions

all any cumsum diag diff end eye find isempty isequal isfinite isinf islogical isnan isnumeric length logical ndims numel ones permute prod reshape size sort sum tril triu zeros True if all elements of a vector are nonzero. True if any element of a vector is nonzero. Cumulative sum of elements. Diagonal matrices and diagonals of a matrix. Difference and approximate derivative. Last index in an indexing expression. Identity matrix. Find indices of nonzero elements. True for empty matrix. True if arrays are numerically equal. True for ﬁnite elements. True for inﬁnite elements. True for logical array. True for Not-a-Number. True for numeric arrays. Length of vector. Convert numeric values to logical. Number of dimensions. Number of elements in a matrix. Ones array. Permute array dimensions. Product of elements. Change size. Size of matrix. Sort in ascending order. Sum of elements. Extract lower triangular part. Extract upper triangular part. Zeros array.

Some of these functions are shorthands for combinations of other built-in functions, lik length(x) ndims(x) numel(x) is is is max(size(x)) length(size(x)) prod(size(x))

Others are shorthands for frequently used tests, like isempty(x) isinf(x) isfinite(x) is is is numel(x) == 0 abs(x) == Inf abs(x) ~= Inf

Others are shorthands for frequently used functions which could have been written with low-level code, like diag, eye, find, sum, cumsum, cumprod, sort, tril, triu, etc.

CHAPTER 2. OPERATORS, FUNCTIONS AND SPECIAL CHARACTERS

5

2.3 M-ﬁle functions

flipdim fliplr flipud ind2sub ipermute kron linspace ndgrid repmat rot90 shiftdim squeeze sub2ind Flip matrix along speciﬁed dimension. Flip matrix in left/right direction. Flip matrix in up/down direction. Multiple subscripts from linear index. Inverse permute array dimensions. Kronecker tensor product. Linearly spaced vector. Generation of arrays for N-D functions and interpolation. Replicate and tile an array. Rotate matrix 90 degrees. Shift dimensions. Remove singleton dimensions. Linear index from multiple subscripts.

Chapter 3

Basic array properties

3.1 Size

The size of an array is a row vector with the length along all dimensions. The size of the array x can be found with

sx = size(x); % size of x (along all dimensions)

The length of the size vector sx is the number of dimensions in x. That is, length(size(x)) is identical to ndims(x) (see section 3.2.1). No builtin array class in M ATLAB has less than two dimensions. To change the size of an array without changing the number of elements, use reshape.

3.1.1 Size along a speciﬁc dimension

To get the length along a speciﬁc dimension dim, of the array x, use

size(x, dim) % size of x (along a specific dimension)

This will return one for all singleton dimensions (see section 3.2.2), and, in particular, it will return one for all dim greater than ndims(x).

3.1.2 Size along multiple dimensions

Sometimes one needs to get the size along multiple dimensions. It would be nice if we could use size(x, dims), where dims is a vector of dimension numbers, but alas, size only allows the dimension argument to be a scalar. We may of course use a for-loop solution like

siz = zeros(size(dims)); for i = 1 : numel(dims) siz(i) = size(x, dims(i)); end % initialize size vector to return % loop over the elements in dims % get the size along dimension % end loop

A vectorized version of the above is

siz = ones(size(dims)); sx = size(x); k = dims <= ndims(x); siz(k) = sx(dims(k)); % % % % initialize size vector to return get size along all dimensions dimensions known not to be trailing singleton insert size along dimensions of interest

6

CHAPTER 3. BASIC ARRAY PROPERTIES

7

which is the essential part of the function mttsize in the MTT Toolbox. Code like the following is sometimes seen, unfortunately. It might be more intuitive than the above, but it is more fragile since it might use a lot of memory when dims contains a large value.

sx = size(x); n = max(dims(:)) - ndims(x); sx = [ sx ones(1, n) ]; siz = sx(dims); % % % % get size along all dimensions number of dimensions to append pad size vector extract dimensions of interest

An unlikely scenario perhaps, but imagine what happens if x and dims both are scalars and that dims is a billion. The above code would require more than 8 GB of memory. The suggested solution further above requires a negligible amount of memory. There is no reason to write fragile code when it can easily be avoided.

3.2 Dimensions

3.2.1 Number of dimensions

The number of dimensions of an array is the number of the highest non-singleton dimension (see section 3.2.2) but never less than two since arrays in M ATLAB always have at least two dimensions. The function which returns the number of dimensions is ndims, so the number of dimensions of an array x is

dx = ndims(x); % number of dimensions

One may also say that ndims(x) is the largest value of dim, no less than two, for which size(x,dim) is different from one. Here are a few examples

x x x x x x = = = = = = ones(2,1) ones(2,1,1,1) ones(1,0) ones(1,2,3,0,0) ones(2,3,0,0,1) ones(3,0,0,1,2) % % % % % % 2-dimensional 2-dimensional 2-dimensional 5-dimensional 4-dimensional 5-dimensional

3.2.2 Singleton dimensions

A “singleton dimension” is a dimension along which the length is one. That is, if size(x,dim) is one, then dim is a singleton dimension. If, in addition, dim is larger than ndims(x), then dim is called a “trailing singleton dimension”. Trailing singleton dimensions are ignored by size and ndims. Singleton dimensions may be removed with squeeze. Removing singleton dimensions changes the size of an array, but it does not change the number of elements in an array Flipping an array along a singleton dimension is a null-operation, that is, it has no effect, it changes nothing.

3.3 Number of elements

The number of elements in an array may be obtained with numel, e.g., numel(x) is the number of elements in x. The number of elements is simply the product of the length along all dimensions, that is, prod(size(x)).

CHAPTER 3. BASIC ARRAY PROPERTIES

8

3.3.1 Empty arrays

If the length along at least one dimension is zero, then the array has zero elements, and hence it is empty. We could test the array x for emptiness with any(size(x) == 0) or numel(x) == 0, but there is a builtin function which explicitly tests for emptiness, isempty.

Chapter 4

Array indices and subscripts

To be written.

9

Chapter 5

Creating basic vectors, matrices and arrays

5.1 Creating a constant array

5.1.1 When the class is determined by the scalar to replicate

To create an array whose size is siz =[m n p q ...] and where each element has the value val, use

X = repmat(val, siz);

Following are three other ways to achieve the same, all based on what repmat uses internally. Note that for these to work, the array X should not already exist

X(prod(siz)) = val; X = reshape(X, siz); X(:) = X(end); % array of right class and num. of elements % reshape to specified size % fill ‘val’ into X (redundant if ‘val’ is zero)

If the size is given as a cell vector siz ={m n p q ...}, there is no need to reshape

X(siz{:}) = val; X(:) = X(end); % array of right class and size % fill ‘val’ into ‘X’ (redundant if ‘val’ is zero)

If m, n, p, q, . . . are scalar variables, one may use

X(m,n,p,q) = val; X(:) = X(end); % array of right class and size % fill ‘val’ into X (redundant if ‘val’ is zero)

The following way of creating a constant array is frequently used

X = val(ones(siz));

but this solution requires more memory since it creates an index array. Since an index array is used, it only works if val is a variable, whereas the other solutions above also work when val is a function returning a scalar value, e.g., if val is Inf or NaN:

X = NaN(ones(siz)); X = repmat(NaN, siz); % this won’t work unless NaN is a variable % here NaN may be a function or a variable

Avoid using 10

CHAPTER 5. CREATING BASIC VECTORS, MATRICES AND ARRAYS

X = val * ones(siz);

11

since it does unnecessary multiplications and only works for classes for which the multiplication operator is deﬁned.

5.1.2 When the class is stored in a string variable

To create an array of an arbitrary class cls, where cls is a character array (i.e., string) containing the class name, use any of the above which allows val to be a function call and let val be

feval(cls, val)

As a special case, to create an array of class cls with only zeros, you can use

X = repmat(feval(cls, 0), siz); % a nice one-liner

or

X(prod(siz)) = feval(cls, 0); X = reshape(X, siz);

Avoid using

X = feval(cls, zeros(siz)); % might require a lot more memory

since it ﬁrst creates an array of class double which might require many times more memory than X if an array of class cls requires less memory pr element than a double array.

5.2 Special vectors

5.2.1 Uniformly spaced elements

To create a vector of uniformly spaced elements, use the linspace function or the : (colon) operator:

X = linspace(lower, upper, n); X = linspace(lower, upper, n).’; X = lower : step : upper; X = ( lower : step : upper ).’; % row vector % column vector % row vector % column vector

If the difference upper-lower is not a multiple of step, the last element of X, X(end), will be less than upper. So the condition A(end) <= upper is always satisﬁed.

Chapter 6

Shifting

6.1 Vectors

To shift and rotate the elements of a vector, use

X([ X([ X([ X([ end 1:end-1 ]); end-k+1:end 1:end-k ]); 2:end 1 ]); k+1:end 1:k ]); % % % % shift shift shift shift right/down 1 element right/down k elements left/up 1 element left/up k elements

Note that these only work if k is non-negative. If k is an arbitrary integer one may use something like

X( mod((1:end)-k-1, end)+1 ); X( mod((1:end)+k-1, end)+1 ); % shift right/down k elements % shift left/up k element

where a negative k will shift in the opposite direction of a positive k.

6.2 Matrices and arrays

To shift and rotate the elements of an array X along dimension dim, ﬁrst initialize a subscript cell array with

idx = repmat({’:’}, ndims(X), 1); % initialize subscripts n = size(X, dim); % length along dimension dim

then manipulate the subscript cell array as appropriate by using one of

idx{dim} idx{dim} idx{dim} idx{dim} = = = = [ [ [ [ n 1:n-1 ]; n-k+1:n 1:n-k ]; 2:n 1 ]; k+1:n 1:k ]; % % % % shift shift shift shift right/down/forwards right/down/forwards left/up/backwards 1 left/up/backwards k 1 element k elements element elements

ﬁnally create the new array

Y = X(idx{:});

12

Chapter 7

Replicating elements and arrays

7.1 Creating a constant array

See section 5.1.

7.2 Replicating elements in vectors

7.2.1 Replicate each element a constant number of times

Example Given

N = 3; A = [ 4 5 ]

create N copies of each element in A, so

B = [ 4 4 4 5 5 5 ]

Use, for instance,

B = A(ones(1,N),:); B = B(:).’;

If A is a column-vector, use

B = A(:,ones(1,N)).’; B = B(:);

Some people use

B = A( ceil( (1:N*length(A))/N ) );

but this requires unnecessary arithmetic. The only advantage is that it works regardless of whether A is a row or column vector.

7.2.2 Replicate each element a variable number of times

See section 15.5.2 about run-length decoding.

13

CHAPTER 7. REPLICATING ELEMENTS AND ARRAYS

14

7.3 Using KRON for replicating elements

7.3.1 KRON with an matrix of ones

Using kron with one of the arguments being a matrix of ones, may be used to replicate elements. Firstly, since the replication is done by multiplying with a matrix of ones, it only works for classes for which the multiplication operator is deﬁned. Secondly, it is never necessary to perform any multiplication to replicate elements. Hence, using kron is not the best way. Assume A is a p-by-q matrix and that n is a non-negative integer. Using KRON with a matrix of ones as ﬁrst argument The expression

B = kron(ones(m,n), A);

may be computed more efﬁciently with

i = (1:p).’; j = (1:q).’; B = A(i,j); i = i(:,ones(1,m)); j = j(:,ones(1,n));

or simply

B = repmat(A, [m n]);

Using KRON with a matrix of ones as second argument The expression

B = kron(A, ones(m,n));

may be computed more efﬁciently with

i = 1:p; i = i(ones(1,m),:); j = 1:q; j = j(ones(1,n),:); B = A(i,j);

7.3.2 KRON with an identity matrix

Assume A is a p-by-q matrix and that n is a non-negative integer. Using KRON with an identity matrix as second argument The expression

B = kron(A, eye(n));

may be computed more efﬁciently with

B = zeros(p, q, n, n); B(:,:,1:n+1:n^2) = repmat(A, [1 1 n]); B = permute(B, [3 1 4 2]); B = reshape(B, [n*p n*q]);

or the following, which does not explicitly use either p or q

B = zeros([size(A) n n]); B(:,:,1:n+1:n^2) = repmat(A, [1 1 n]); B = permute(B, [3 1 4 2]); B = reshape(B, n*size(A));

CHAPTER 7. REPLICATING ELEMENTS AND ARRAYS

Using KRON with an identity matrix as ﬁrst argument The expression

B = kron(eye(n), A);

15

may be computed more efﬁciently with

B = zeros(p, q, n, n); B(:,:,1:n+1:n^2) = repmat(A, [1 1 n]); B = permute(B, [1 3 2 4]); B = reshape(B, [n*p n*q]);

or the following, which does not explicitly use either p or q

B = zeros([size(A) n n]); B(:,:,1:n+1:n^2) = repmat(A, [1 1 n]); B = permute(B, [1 3 2 4]); B = reshape(B, n*size(A));

Chapter 8

Reshaping arrays

8.1 Subdividing 2D matrix

Assume X is an m-by-n matrix.

8.1.1 Create 4D array

To create a p-by-q-by-m/p-by-n/q array Y where the i,j submatrix of X is Y(:,:,i,j), use

Y = reshape( X, [ p m/p q n/q ] ); Y = permute( Y, [ 1 3 2 4 ] );

Now,

X = [ Y(:,:,1,1) Y(:,:,2,1) ... Y(:,:,m/p,1) Y(:,:,1,2) ... Y(:,:,1,n/q) Y(:,:,2,2) ... Y(:,:,2,n/q) ... ... ... Y(:,:,m/p,2) ... Y(:,:,m/p,n/q) ];

To restore X from Y use

X = permute( Y, [ 1 3 2 4 ] ); X = reshape( X, [ m n ] );

8.1.2 Create 3D array (columns ﬁrst)

Assume you want to create a p-by-q-by-m*n/(p*q) array Y where the i,j submatrix of X is Y(:,:,i+(j-1)*m/p). E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = cat( 3, A, C, B, D );

use

Y = reshape( X, [ p m/p q n/q ] ); Y = permute( Y, [ 1 3 2 4 ] ); Y = reshape( Y, [ p q m*n/(p*q) ] )

16

CHAPTER 8. RESHAPING ARRAYS

Now,

X = [ Y(:,:,1) Y(:,:,m/p+1) ... Y(:,:,(n/q-1)*m/p+1) Y(:,:,2) Y(:,:,m/p+2) ... Y(:,:,(n/q-1)*m/p+2) ... ... ... ... Y(:,:,m/p) Y(:,:,2*m/p) ... Y(:,:,n/q*m/p)

17

];

To restore X from Y use

X = reshape( Y, [ p q m/p n/q ] ); X = permute( X, [ 1 3 2 4 ] ); X = reshape( X, [ m n ] );

8.1.3 Create 3D array (rows ﬁrst)

Assume you want to create a p-by-q-by-m*n/(p*q) array Y where the i,j submatrix of X is Y(:,:,j+(i-1)*n/q). E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = cat( 3, A, B, C, D );

use

Y = reshape( X, [ p m/p n ] ); Y = permute( Y, [ 1 3 2 ] ); Y = reshape( Y, [ p q m*n/(p*q) ] );

Now,

X = [ Y(:,:,1) Y(:,:,2) Y(:,:,n/q+1) Y(:,:,n/q+2) ... ... Y(:,:,(m/p-1)*n/q+1) Y(:,:,(m/p-1)*n/q+2) ... Y(:,:,n/q) ... Y(:,:,2*n/q) ... ... ... Y(:,:,m/p*n/q) ];

To restore X from Y use

X = reshape( Y, [ p n m/p ] ); X = permute( X, [ 1 3 2 ] ); X = reshape( X, [ m n ] );

8.1.4 Create 2D matrix (columns ﬁrst, column output)

Assume you want to create a m*n/q-by-q matrix Y where the submatrices of X are concatenated (columns ﬁrst) vertically. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = [ A C B D ];

CHAPTER 8. RESHAPING ARRAYS

use

Y = reshape( X, [ m q n/q ] ); Y = permute( Y, [ 1 3 2 ] ); Y = reshape( Y, [ m*n/q q ] );

18

To restore X from Y use

X = reshape( Y, [ m n/q q ] ); X = permute( X, [ 1 3 2 ] ); X = reshape( X, [ m n ] );

8.1.5 Create 2D matrix (columns ﬁrst, row output)

Assume you want to create a p-by-m*n/p matrix Y where the submatrices of X are concatenated (columns ﬁrst) horizontally. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = [ A C B D ];

use

Y = reshape( X, [ p m/p q n/q ] ) Y = permute( Y, [ 1 3 2 4 ] ); Y = reshape( Y, [ p m*n/p ] );

To restore X from Y use

Z = reshape( Y, [ p q m/p n/q ] ); Z = permute( Z, [ 1 3 2 4 ] ); Z = reshape( Z, [ m n ] );

8.1.6 Create 2D matrix (rows ﬁrst, column output)

Assume you want to create a m*n/q-by-q matrix Y where the submatrices of X are concatenated (rows ﬁrst) vertically. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = [ A B C D ];

use

Y = reshape( X, [ p m/p q n/q ] ); Y = permute( Y, [ 1 4 2 3 ] ); Y = reshape( Y, [ m*n/q q ] );

To restore X from Y use

X = reshape( Y, [ p n/q m/p q ] ); X = permute( X, [ 1 3 4 2 ] ); X = reshape( X, [ m n ] );

CHAPTER 8. RESHAPING ARRAYS

19

8.1.7 Create 2D matrix (rows ﬁrst, row output)

Assume you want to create a p-by-m*n/p matrix Y where the submatrices of X are concatenated (rows ﬁrst) horizontally. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B C D ];

into

Y = [ A B C D ];

use

Y = reshape( X, [ p m/p n ] ); Y = permute( Y, [ 1 3 2 ] ); Y = reshape( Y, [ p m*n/p ] );

To restore X from Y use

X = reshape( Y, [ p n m/p ] ); X = permute( X, [ 1 3 2 ] ); X = reshape( X, [ m n ] );

8.2 Stacking and unstacking pages

Assume X is a m-by-n-by-p array and you want to create an m*p-by-n matrix Y that contains the pages of X stacked vertically. E.g., if A, B, C, etc. are m-by-n matrices, then, to convert

X = cat(3, A, B, C, ...);

into

Y = [ A B C ... ];

use

Y = permute( X, [ 1 3 2 ] ); Y = reshape( Y, [ m*p n ] );

To restore X from Y use

X = reshape( Y, [ m p n ] ); X = permute( X, [ 1 3 2 ] );

Chapter 9

Rotating matrices and arrays

9.1 Rotating 2D matrices

To rotate an m-by-n matrix X, k times 90° counterclockwise one may use

Y = rot90(X, k);

or one may do it like this

Y = X(:,n:-1:1).’; Y = X(m:-1:1,:).’; Y = X(m:-1:1,n:-1:1); % rotate 90 degrees counterclockwise % rotate 90 degrees clockwise % rotate 180 degrees

In the above, one may replace m and n with end.

9.2 Rotating ND arrays

Assume X is an ND array and one wants the rotation to be vectorized along higher dimensions. That is, the same rotation should be performed on all 2D slices X(:,:,i,j,...). Rotating 90 degrees counterclockwise

s v Y Y = = = = size(X); % size vector [ 2 1 3:ndims(X) ]; % dimension permutation vector permute( X(:,s(2):-1:1,:), v ); reshape( Y, s(v) );

Rotating 180 degrees

s = size(X); Y = reshape( X(s(1):-1:1,s(2):-1:1,:), s );

or the one-liner

Y = reshape( X(end:-1:1,end:-1:1,:), size(X) );

20

CHAPTER 9. ROTATING MATRICES AND ARRAYS

Rotating 90 clockwise

s v Y Y = = = = size(X); % size vector [ 2 1 3:ndims(X) ]; % dimension permutation vector reshape( X(s(1):-1:1,:), s ); permute( Y, v );

21

or the one-liner

Y = permute(reshape(X(end:-1:1,:), size(X)), [2 1 3:ndims(X)]);

9.3 Rotating ND arrays around an arbitrary axis

When rotating an ND array X we need to specify the axis around which the rotation should be performed. The general case is to rotate an array around an axis perpendicular to the plane spanned by dim1 and dim2. In the cases above, the rotation was performed around an axis perpendicular to a plane spanned by dimensions one (rows) and two (columns). Note that a rotation changes nothing if both size(X,dim1) and size(X,dim2) is one.

% Largest dimension number we have to deal with. nd = max( [ ndims(X) dim1 dim2 ] ); % Initialize subscript cell array. v = repmat({’:’}, [nd 1]);

then, depending on how to rotate, use Rotate 90 degrees counterclockwise

v{dim2} = size(X,dim2):-1:1; Y = X(v{:}); d = 1:nd; d([ dim1 dim2 ]) = [ dim2 dim1 ]; Y = permute(X, d);

Rotate 180 degrees

v{dim1} = size(X,dim1):-1:1; v{dim2} = size(X,dim2):-1:1; Y = X(v{:});

Rotate 90 degrees clockwise

v{dim1} = size(X,dim1):-1:1; Y = X(v{:}); d = 1:nd; d([ dim1 dim2 ]) = [ dim2 dim1 ]; Y = permute(X, d);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

22

9.4 Block-rotating 2D matrices

9.4.1 “Inner” vs “outer” block rotation

When talking about block-rotation of arrays, we have to differentiate between two different kinds of rotation. Lacking a better name I chose to call it “inner block rotation” and “outer block rotation”. Inner block rotation is a rotation of the elements within each block, preserving the position of each block within the array. Outer block rotation rotates the blocks but does not change the position of the elements within each block. An example will illustrate: An inner block rotation 90 degrees counterclockwise will have the following effect

[ A B C D E F G H I ] => [ rot90(A) rot90(B) rot90(C) rot90(D) rot90(E) rot90(F) rot90(G) rot90(H) rot90(I) ]

However, an outer block rotation 90 degrees counterclockwise will have the following effect

[ A B C D E F G H I ] => [ C F I B E H A D G ]

In all the examples below, it is assumed that X is an m-by-n matrix of p-by-q blocks.

CHAPTER 9. ROTATING MATRICES AND ARRAYS

23

9.4.2 “Inner” block rotation 90 degrees counterclockwise

General case To perform the rotation

X = [ A B ... C D ... ... ... ] => [ rot90(A) rot90(B) ... rot90(C) rot90(D) ... ... ... ... ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q n/q ] ); Y(:,:,q:-1:1,:); permute( Y, [ 3 2 1 4 ] ); reshape( Y, [ q*m/p p*n/q ] ); % or Y = Y(:,:,end:-1:1,:);

Special case: m=p To perform the rotation

[ A B ... ] => [ rot90(A) rot90(B) ... ]

use

Y Y Y Y = = = = reshape( X, [ p q n/q ] ); Y(:,q:-1:1,:); permute( Y, [ 2 1 3 ] ); reshape( Y, [ q m*n/q ] ); % or Y = Y(:,end:-1:1,:); % or Y = Y(:,:);

Special case: n=q To perform the rotation

X = [ A B ... ] => [ rot90(A) rot90(B) ... ]

use

Y Y Y Y = = = = X(:,q:-1:1); reshape( Y, [ p m/p q ] ); permute( Y, [ 3 2 1 ] ); reshape( Y, [ q*m/p p ] ); % or Y = X(:,end:-1:1);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

24

9.4.3 “Inner” block rotation 180 degrees

General case To perform the rotation

X = [ A B ... C D ... ... ... ] => [ rot90(A,2) rot90(B,2) ... rot90(C,2) rot90(D,2) ... ... ... ... ]

use

Y = reshape( X, [ p m/p q n/q ] ); Y = Y(p:-1:1,:,q:-1:1,:); Y = reshape( Y, [ m n ] ); % or Y = Y(end:-1:1,:,end:-1:1,:);

Special case: m=p To perform the rotation

[ A B ... ] => [ rot90(A,2) rot90(B,2) ... ]

use

Y = reshape( X, [ p q n/q ] ); Y = Y(p:-1:1,q:-1:1,:); Y = reshape( Y, [ m n ] ); % or Y = Y(end:-1:1,end:-1:1,:); % or Y = Y(:,:);

Special case: n=q To perform the rotation

X = [ A B ... ] => [ rot90(A,2) rot90(B,2) ... ]

use

Y = reshape( X, [ p m/p q ] ); Y = Y(p:-1:1,:,q:-1:1); Y = reshape( Y, [ m n ] ); % or Y = Y(end:-1:1,:,end:-1:1);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

25

9.4.4 “Inner” block rotation 90 degrees clockwise

General case To perform the rotation

X = [ A B ... C D ... ... ... ] => [ rot90(A,3) rot90(B,3) ... rot90(C,3) rot90(D,3) ... ... ... ... ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q n/q ] ); Y(p:-1:1,:,:,:); permute( Y, [ 3 2 1 4 ] ); reshape( Y, [ q*m/p p*n/q ] ); % or Y = Y(end:-1:1,:,:,:);

Special case: m=p To perform the rotation

[ A B ... ] => [ rot90(A,3) rot90(B,3) ... ]

use

Y Y Y Y = = = = X(p:-1:1,:); reshape( Y, [ p q n/q ] ); permute( Y, [ 2 1 3 ] ); reshape( Y, [ q m*n/q ] ); % or Y = X(end:-1:1,:);

% or Y = Y(:,:);

Special case: n=q To perform the rotation

X = [ A B ... ] => [ rot90(A,3) rot90(B,3) ... ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q ] ); Y(p:-1:1,:,:); permute( Y, [ 3 2 1 ] ); reshape( Y, [ q*m/p p ] ); % or Y = Y(end:-1:1,:,:);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

26

9.4.5 “Outer” block rotation 90 degrees counterclockwise

General case To perform the rotation

X = [ A B ... C D ... ... ... ] => [ ... ... B D ... A C ... ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q n/q ] ); Y(:,:,:,n/q:-1:1); permute( Y, [ 1 4 3 2 ] ); reshape( Y, [ p*n/q q*m/p ] ); % or Y = Y(:,:,:,end:-1:1);

Special case: m=p To perform the rotation

[ A B ... ] => [ ... B A ]

use

Y Y Y Y = = = = reshape( X, [ p q n/q ] ); Y(:,:,n/q:-1:1); permute( Y, [ 1 3 2 ] ); reshape( Y, [ m*n/q q ] ); % or Y = Y(:,:,end:-1:1);

Special case: n=q To perform the rotation

X = [ A B ... ] => [ A B ... ]

use

Y = reshape( X, [ p m/p q ] ); Y = permute( Y, [ 1 3 2 ] ); Y = reshape( Y, [ p n*m/p ] );

% or Y(:,:);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

27

9.4.6 “Outer” block rotation 180 degrees

General case To perform the rotation

X = [ A B ... C D ... ... ... ] [ => ... ... ... D C ... B A ]

use

Y = reshape( X, [ p m/p q n/q ] ); Y = Y(:,m/p:-1:1,:,n/q:-1:1); % or Y = Y(:,end:-1:1,:,end:-1:1); Y = reshape( Y, [ m n ] );

Special case: m=p To perform the rotation

[ A B ... ] => [ ... B A ]

use

Y = reshape( X, [ p q n/q ] ); Y = Y(:,:,n/q:-1:1); Y = reshape( Y, [ m n ] ); % or Y = Y(:,:,end:-1:1); % or Y = Y(:,:);

Special case: n=q To perform the rotation

X = [ A B ... ] => [ ... B A ]

use

Y = reshape( X, [ p m/p q ] ); Y = Y(:,m/p:-1:1,:); Y = reshape( Y, [ m n ] ); % or Y = Y(:,end:-1:1,:);

CHAPTER 9. ROTATING MATRICES AND ARRAYS

28

9.4.7 “Outer” block rotation 90 degrees clockwise

General case To perform the rotation

X = [ A B ... C D ... ... ... ] => [ ... C A ... D B ... ... ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q n/q ] ); Y(:,m/p:-1:1,:,:); permute( Y, [ 1 4 3 2 ] ); reshape( Y, [ p*n/q q*m/p ] ); % or Y = Y(:,end:-1:1,:,:);

Special case: m=p To perform the rotation

[ A B ... ] => [ A B ... ]

use

Y = reshape( X, [ p q n/q ] ); Y = permute( Y, [ 1 3 2 ] ); Y = reshape( Y, [ m*n/q q ] );

Special case: n=q To perform the rotation

X = [ A B ... ] => [ ... B A ]

use

Y Y Y Y = = = = reshape( X, [ p m/p q ] ); Y(:,m/p:-1:1,:); permute( Y, [ 1 3 2 ] ); reshape( Y, [ p n*m/p ] ); % or Y = Y(:,end:-1:1,:);

9.5 Blocktransposing a 2D matrix

9.5.1 “Inner” blocktransposing

Assume X is an m-by-n matrix and you want to subdivide it into p-by-q submatrices and transpose as if each block was an element. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B ... C D ... ... ... ] => [ A.’ B.’ ... C.’ D.’ ... ... ... ... ]

use

Y = reshape( X, [ p m/p q n/q ] ); Y = permute( Y, [ 3 2 1 4 ] ); Y = reshape( Y, [ q*m/p p*n/q ] );

CHAPTER 9. ROTATING MATRICES AND ARRAYS

29

9.5.2 “Outer” blocktransposing

Assume X is an m-by-n matrix and you want to subdivide it into p-by-q submatrices and transpose as if each block was an element. E.g., if A, B, C and D are p-by-q matrices, convert

X = [ A B ... C D ... ... ... ] => [ A C ... B D ... ... ... ]

use

Y = reshape( X, [ p m/p q n/q ] ); Y = permute( Y, [ 1 4 3 2 ] ); Y = reshape( Y, [ p*n/q q*m/p] );

Chapter 10

Basic arithmetic operations

10.1 Multiply arrays

10.1.1 Multiply each 2D slice with the same matrix (element-by-element)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is an m-by-n matrix and you want to construct a new m-by-n-by-p-by-q-by-. . . array Z, where

Z(:,:,i,j,...) = X(:,:,i,j,...) .* Y;

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by the following vectorized code

sx = size(X); Z = X .* repmat(Y, [1 1 sx(3:end)]);

10.1.2 Multiply each 2D slice with the same matrix (left)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is a k-by-m matrix and you want to construct a new k-by-n-by-p-by-q-by-. . . array Z, where

Z(:,:,i,j,...) = Y * X(:,:,i,j,...);

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by the following vectorized code

sx = size(X); sy = size(Y); Z = reshape(Y * X(:,:), [sy(1) sx(2:end)]);

The above works by reshaping X so that all 2D slices X(:,:,i,j,...) are placed next to each other (horizontal concatenation), then multiply with Y, and then reshaping back again. The X(:,:) is simply a short-hand for reshape(X, [sx(1) prod(sx)/sx(1)]).

10.1.3 Multiply each 2D slice with the same matrix (right)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is an n-by-k matrix and you want to construct a new m-by-n-by-p-by-q-by-. . . array Z, where 30

CHAPTER 10. BASIC ARITHMETIC OPERATIONS

Z(:,:,i,j,...) = X(:,:,i,j,...) * Y;

31

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by vectorized code. First create the variables

sx = size(X); sy = size(Y); dx = ndims(X);

Then use the fact that

Z(:,:,i,j,...) = X(:,:,i,j,...) * Y = (Y’ * X(:,:,i,j,...)’)’;

so the multiplication Y’ * X(:,:,i,j,...)’ can be solved by the method in section 10.1.2.

Xt = conj(permute(X, [2 1 3:dx])); Z = Y’ * Xt(:,:); Z = reshape(Z, [sy(2) sx(1) sx(3:dx)]); Z = conj(permute(Z, [2 1 3:dx]));

Note how the complex conjugate transpose (’) on the 2D slices of X was replaced by a combination of permute and conj. Actually, because signs will cancel each other, we can simplify the above by removing the calls to conj and replacing the complex conjugate transpose (’) with the non-conjugate transpose (.’). The code above then becomes

Xt = permute(X, [2 1 3:dx]); Z = Y.’ * Xt(:,:); Z = reshape(Z, [sy(2) sx(1) sx(3:dx)]); Z = permute(Z, [2 1 3:dx]);

An alternative method is to perform the multiplication X(:,:,i,j,...) * Y directly but that requires that we stack all 2D slices X(:,:,i,j,...) on top of each other (vertical concatenation), multiply, and unstack. The code is then

Xt = permute(X, [1 3:dx 2]); Xt = reshape(Xt, [prod(sx)/sx(2) sx(2)]); Z = Xt * Y; Z = reshape(Z, [sx(1) sx(3:dx) sy(2)]); Z = permute(Z, [1 dx 2:dx-1]);

The ﬁrst two lines perform the stacking and the two last perform the unstacking.

10.1.4 Multiply matrix with every element of a vector

Assume X is an m-by-n matrix and v is a vector with length p. How does one write

Y = zeros(m, n, p); for i = 1:p Y(:,:,i) = X * v(i); end

with no for-loop? One way is to use

Y = reshape(X(:)*v, [m n p]);

For the more general problem where X is an m-by-n-by-p-by-q-by-... array and v is a p-by-qby-... array, the for-loop

CHAPTER 10. BASIC ARITHMETIC OPERATIONS

Y = zeros(m, n, p, q, ...); ... for j = 1:q for i = 1:p Y(:,:,i,j,...) = X(:,:,i,j,...) * v(i,j,...); end end ...

32

may be written as

sx = size(X); Z = X .* repmat(reshape(v, [1 1 sx(3:end)]), [sx(1) sx(2)]);

10.1.5 Multiply each 2D slice with corresponding element of a vector

Assume X is an m-by-n-by-p array and v is a row vector with length p. How does one write

Y = zeros(m, n, p); for i = 1:p Y(:,:,i) = X(:,:,i) * v(i); end

with no for-loop? One way is to use

Y = X .* repmat(reshape(v, [1 1 p]), [m n]);

10.1.6 Outer product of all rows in a matrix

Assume X is an m-by-n matrix. How does one create an n-by-n-by-m matrix Y so that, for all i from 1 to m,

Y(:,:,i) = X(i,:)’ * X(i,:);

The obvious for-loop solution is

Y = zeros(n, n, m); for i = 1:m Y(:,:,i) = X(i,:)’ * X(i,:); end

a non-for-loop solution is

j = 1:n; Y = reshape(repmat(X’, n, 1) .* X(:,j(ones(n, 1),:)).’, [n n m]);

Note the use of the non-conjugate transpose in the second factor to ensure that it works correctly also for complex matrices.

10.1.7 Keeping only diagonal elements of multiplication

Assume X and Y are two m-by-n matrices and that W is an n-by-n matrix. How does one vectorize the following for-loop

CHAPTER 10. BASIC ARITHMETIC OPERATIONS

Z = zeros(m, 1); for i = 1:m Z(i) = X(i,:)*W*Y(i,:)’; end

33

Two solutions are

Z = diag(X*W*Y’); Z = sum(X*W.*conj(Y), 2); % (1) % (2)

Solution (1) does a lot of unnecessary work, since we only keep the n diagonal elements of the nˆ2 computed elements. Solution (2) only computes the elements of interest and is signiﬁcantly faster if n is large.

10.1.8 Products involving the Kronecker product

The following is based on a posting by Paul Fackler <paul_fackler@ncsu.edu> to the Usenet news group comp.soft-sys.matlab. Kronecker products of the form kron(A, eye(n)) are often used to premultiply (or postmultiply) another matrix. If this is the case it is not necessary to actually compute and store the Kronecker product. Assume A is an p-by-q matrix and that B is a q*n-by-m matrix. Then the following two p*n-by-m matrices are identical

C1 = kron(A, eye(n))*B; C2 = reshape(reshape(B.’, [n*m q])*A.’, [m p*n]).’;

The following two p*n-by-m matrices are also identical.

C1 = kron(eye(n), A)*B; C2 = reshape(A*reshape(B, [q n*m]), [p*n m]);

10.2 Divide arrays

10.2.1 Divide each 2D slice with the same matrix (element-by-element)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is an m-by-n matrix and you want to construct a new m-by-n-by-p-by-q-by-. . . array Z, where

Z(:,:,i,j,...) = X(:,:,i,j,...) ./ Y;

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by the following vectorized code

sx = size(X); Z = X./repmat(Y, [1 1 sx(3:end)]);

10.2.2 Divide each 2D slice with the same matrix (left)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is an m-by-m matrix and you want to construct a new m-by-n-by-p-by-q-by-. . . array Z, where

Z(:,:,i,j,...) = Y \ X(:,:,i,j,...);

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by the following vectorized code

Z = reshape(Y\X(:,:), size(X));

CHAPTER 10. BASIC ARITHMETIC OPERATIONS

34

10.2.3 Divide each 2D slice with the same matrix (right)

Assume X is an m-by-n-by-p-by-q-by-. . . array and Y is an m-by-m matrix and you want to construct a new m-by-n-by-p-by-q-by-. . . array Z, where

Z(:,:,i,j,...) = X(:,:,i,j,...) / Y;

for all i=1,...,p, j=1,...,q, etc. This can be done with nested for-loops, or by the following vectorized code

sx = size(X); dx = ndims(X); Xt = reshape(permute(X, [1 3:dx 2]), [prod(sx)/sx(2) sx(2)]); Z = Xt/Y; Z = permute(reshape(Z, sx([1 3:dx 2])), [1 dx 2:dx-1]);

The third line above builds a 2D matrix which is a vertical concatenation (stacking) of all 2D slices X(:,:,i,j,...). The fourth line does the actual division. The ﬁfth line does the opposite of the third line. The ﬁve lines above might be simpliﬁed a little by introducing a dimension permutation vector

sx = size(X); dx = ndims(X); v = [1 3:dx 2]; Xt = reshape(permute(X, v), [prod(sx)/sx(2) sx(2)]); Z = Xt/Y; Z = ipermute(reshape(Z, sx(v)), v);

If you don’t care about readability, this code may also be written as

sx = size(X); dx = ndims(X); v = [1 3:dx 2]; Z = ipermute(reshape(reshape(permute(X, v), ... [prod(sx)/sx(2) sx(2)])/Y, sx(v)), v);

Chapter 11

More complicated arithmetic operations

11.1 Calculating distances

11.1.1 Euclidean distance

The Euclidean distance from xi to y j is di j = x i − y j = (x1i − y1 j )2 + · · · + (x pi − y p j )2

11.1.2 Distance between two points

To calculate the Euclidean distance from a point represented by the vector x to another point represeted by the vector y, use one of

d = norm(x-y); d = sqrt(sum(abs(x-y).^2));

11.1.3 Euclidean distance vector

Assume X is an m-by-p matrix representing m points in p-dimensional space and y is a 1-by-p vector representing a single point in the same space. Then, to compute the m-by-1 distance vector d where d(i) is the Euclidean distance between X(i,:) and y, use

d = sqrt(sum(abs(X - repmat(y, [m 1])).^2, 2)); d = sqrt(sum(abs(X - y(ones(m,1),:)).^2, 2)); % inline call to repmat

11.1.4 Euclidean distance matrix

Assume X is an m-by-p matrix representing m points in p-dimensional space and Y is an n-by-p matrix representing another set of points in the same space. Then, to compute the m-by-n distance matrix D where D(i,j) is the Euclidean distance X(i,:) between Y(j,:), use

D = sqrt(sum(abs( repmat(permute(X, [1 3 2]), [1 n 1]) ... - repmat(permute(Y, [3 1 2]), [m 1 1]) ).^2, 3));

35

CHAPTER 11. MORE COMPLICATED ARITHMETIC OPERATIONS

36

The following code inlines the call to repmat, but requires to temporary variables unless one doesn’t mind changing X and Y

Xt = permute(X, [1 3 2]); Yt = permute(Y, [3 1 2]); D = sqrt(sum(abs( Xt(:,ones(1,n),:) ... - Yt(ones(1,m),:,:) ).^2, 3));

The distance matrix may also be calculated without the use of a 3-D array:

i = (1:m).’; % index vector for x i = i(:,ones(1,n)); % index matrix for x j = 1:n; % index vector for y j = j(ones(1,m),:); % index matrix for y D = zeros(m, n); % initialise output matrix D(:) = sqrt(sum(abs(X(i(:),:) - Y(j(:),:)).^2, 2));

11.1.5 Special case when both matrices are identical

If X and Y are identical one may use the following, which is nothing but a rewrite of the code above

D = sqrt(sum(abs( repmat(permute(X, [1 3 2]), [1 m 1]) ... - repmat(permute(X, [3 1 2]), [m 1 1]) ).^2, 3));

One might want to take advantage of the fact that D will be symmetric. The following code ﬁrst creates the indices for the upper triangular part of D. Then it computes the upper triangular part of D and ﬁnally lets the lower triangular part of D be a mirror image of the upper triangular part.

[ i j ] = find(triu(ones(m), 1)); % trick to get indices D = zeros(m, m); % initialise output matrix D( i + m*(j-1) ) = sqrt(sum(abs( X(i,:) - X(j,:) ).^2, 2)); D( j + m*(i-1) ) = D( i + m*(j-1) );

11.1.6 Mahalanobis distance

¯ The Mahalanobis distance from a vector y j to the set X = {x1 , . . . , xnx } is the distance from y j to x, the centroid of X , weighted according to Cx , the variance matrix of the set X . I.e., ¯ ¯ d2 = (y j − x) Cx −1 (y j − x) j where ¯ x= 1 n ∑ xi nx i=1 and Cx = 1 nx ¯ ¯ ∑ (xi − x)(xi − x) nx − 1 i=1

Assume Y is an ny-by-p matrix containing a set of vectors and X is an nx-by-p matrix containing another set of vectors, then the Mahalanobis distance from each vector Y(j,:) (for j=1,...,ny) to the set of vectors in X can be calculated with

nx = size(X, 1); % size of set in X ny = size(Y, 1); % size of set in Y m = mean(X); C = cov(X); d = zeros(ny, 1); for j = 1:ny d(j) = (Y(j,:) - m) / C * (Y(j,:) - m)’; end

CHAPTER 11. MORE COMPLICATED ARITHMETIC OPERATIONS

37

which is computed more efﬁciently with the following code which does some inlining of functions (mean and cov) and vectorization

nx = size(X, 1); ny = size(Y, 1); m Xc C Yc d = = = = = sum(X, 1)/nx; X - m(ones(nx,1),:); (Xc’ * Xc)/(nx - 1); Y - m(ones(ny,1),:); sum(Yc/C.*Yc, 2)); % size of set in X % size of set in Y % % % % % centroid (mean) distance to centroid of X variance matrix distance to centroid of X Mahalanobis distances

In the complex case, the last line has to be written as

d = real(sum(Yc/C.*conj(Yc), 2)); % Mahalanobis distances

The call to conj is to make sure it also works for the complex case. The call to real is to remove “numerical noise”. The Statistics Toolbox contains the function mahal for calculating the Mahalanobis distances, but mahal computes the distances by doing an orthogonal-triangular (QR) decomposition of the matrix C. The code above returns the same as d = mahal(Y, X). Special case when both matrices are identical If Y and X are identical in the code above, the code may be simpliﬁed somewhat. The for-loop solution becomes

n = size(X, 1); % size of set in X m = mean(X); C = cov(X); d = zeros(n, 1); for j = 1:n d(j) = (Y(j,:) - m) / C * (Y(j,:) - m)’; end

which is computed more efﬁciently with

n m Xc C d = = = = = size(x, 1); sum(x, 1)/n; x - m(ones(n,1),:); (Xc’ * Xc)/(n - 1); sum(Xc/C.*Xc, 2); % % % % centroid (mean) distance to centroid of X variance matrix Mahalanobis distances

Again, to make it work in the complex case, the last line must be written as

d = real(sum(Xc/C.*conj(Xc), 2)); % Mahalanobis distances

Chapter 12

Statistics, probability and combinatorics

12.1 Discrete uniform sampling with replacement

To generate an array X with size vector s, where X contains a random sample from the numbers 1,...,n use

X = ceil(n*rand(s));

To generate a sample from the numbers a,...,b use

X = a + floor((b-a+1)*rand(s));

12.2 Discrete weighted sampling with replacement

Assume p is a vector of probabilities that sum up to 1. Then, to generate an array X with size vector s, where the probability of X(i) being i is p(i) use

m = c = R = X = for X = end length(p); cumsum(p); rand(s); ones(s); i = 1:m-1 X + (R > c(i)); % number of probabilities % cumulative sum

Note that the number of times through the loop depends on the number of probabilities and not the sample size, so it should be quite fast even for large samples.

12.3 Discrete uniform sampling without replacement

To generate a sample of size k from the integers 1,...,n, one may use

X = randperm(n); x = X(1:k);

although that method is only practical if N is reasonably small. 38

CHAPTER 12. STATISTICS, PROBABILITY AND COMBINATORICS

39

12.4 Combinations

“Combinations” is what you get when you pick k elements, without replacement, from a sample of size n, and consider the order of the elements to be irrelevant.

12.4.1 Counting combinations

The number of ways to pick k elements, without replacement, from a sample of size n is is calculated with

c = nchoosek(n, k);

n k

which

one may also use the deﬁnition directly

k = min(k, n-k); % use symmetry property c = round(prod( ((n-k+1):n) ./ (1:k) ));

which is safer than using

k = min(k, n-k); % use symmetry property c = round( prod((n-k+1):n) / prod(1:k) );

which may overﬂow. Unfortunately, both n and k have to be scalars. If n and/or k are vectors, one may use the fact that n! n Γ(n + 1) = = k! (n − k)! Γ(k + 1) Γ(n − k + 1) k and calculate this in with

round(exp(gammaln(n+1) - gammaln(k+1) - gammaln(n-k+1)))

where the round is just to remove any “numerical noise” that might have been introduced by gammaln and exp.

12.4.2 Generating combinations

To generate a matrix with all possible combinations of n elements taken k at a time, one may use the M ATLAB function nchoosek. That function is rather slow compared to the choosenk function which is a part of Mike Brookes’ Voicebox (Speech recognition toolbox) whose homepage is http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html For the special case of generating all combinations of n elements taken 2 at a time, there is a neat trick

[ x(:,2) x(:,1) ] = find(tril(ones(n), -1));

12.5 Permutations

12.5.1 Counting permutations

p = prod(n-k+1:n);

CHAPTER 12. STATISTICS, PROBABILITY AND COMBINATORICS

40

12.5.2 Generating permutations

To generate a matrix with all possible permutations of n elements, one may use the function perms. That function is rather slow compared to the permutes function which is a part of Mike Brookes’ Voicebox (Speech recognition toolbox) whose homepage is at http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html

Chapter 13

Identifying types of arrays

13.1 Numeric array

A numeric array is an array that contains real or complex numerical values including NaN and Inf. An array is numeric if its class is double, single, uint8, uint16, uint32, int8, int16 or int32. To see if an array x is numeric, use

isnumeric(x)

To disallow NaN and Inf, we can not just use

isnumeric(x) & ~any(isnan(x(:))) & ~any(isinf(x(:)))

since, by default, isnan and isinf are only deﬁned for class double. A solution that works is to use the following, where tf is either true or false

tf = isnumeric(x); if isa(x, ’double’) tf = tf & ~any(isnan(x(:))) & ~any(isinf(x(:))) end

If one is only interested in arrays of class double, the above may be written as

isa(x,’double’) & ~any(isnan(x(:))) & ~any(isinf(x(:)))

Note that there is no need to call isnumeric in the above, since a double array is always numeric.

13.2 Real array

M ATLAB has a subtle distinction between arrays that have a zero imaginary part and arrays that do not have an imaginary part:

isreal(0) isreal(complex(0, 0)) % no imaginary part, so true % imaginary part (which is zero), so false

The essence is that isreal returns false (i.e., 0) if space has been allocated for an imaginary part. It doesn’t care if the imaginary part is zero, if it is present, then isreal returns false. To see if an array x is real in the sense that it has no non-zero imaginary part, use

~any(imag(x(:)))

Note that x might be real without being numeric; for instance, isreal(’a’) returns true, but isnumeric(’a’) returns false. 41

CHAPTER 13. IDENTIFYING TYPES OF ARRAYS

42

13.3 Identify real or purely imaginary elements

To see which elements are real or purely imaginary, use

imag(x) == 0 ~imag(x) real(x) ~= 0 logical(real(x)) % identify real elements % ditto (might be faster) % identify purely imaginary elements % ditto (might be faster)

13.4 Array of negative, non-negative or positive values

To see if the elements of the real part of all elements of x are negative, non-negative or positive values, use

x < 0 all(x(:) < 0) x >= 0 all(x(:) >= 0) x > 0 all(x(:) > 0) % identify negative elements % see if all elements are negative % identify non-negative elements % see if all elements are non-negative % identify positive elements % see if all elements are positive

13.5 Array of integers

To see if an array x contains real or complex integers, use

x == round(x) ~imag(x) & x == round(x) % identify (possibly complex) integers % identify real integers

% see if x contains only (possibly complex) integers all(x(:) == round(x(:))) % see if x contains only real integers isreal(x) & all(x(:) == round(x(:)))

13.6 Scalar

To see if an array x is scalar, i.e., an array with exactly one element, use

all(size(x) == 1) prod(size(x)) == 1 any(size(x) ~= 1) prod(size(x)) ~= 1 % % % % is is is is a scalar a scalar not a scalar not a scalar

An array x is scalar or empty if the following is true

isempty(x) | all(size(x) == 1) prod(size(x)) <= 1 prod(size(x)) > 1 % is scalar or empty % is scalar or empty % is not scalar or empty

CHAPTER 13. IDENTIFYING TYPES OF ARRAYS

43

13.7 Vector

An array x is a non-empty vector if the following is true

~isempty(x) & sum(size(x) > 1) <= 1 isempty(x) | sum(size(x) > 1) > 1 % is a non-empty vector % is not a non-empty vector

An array x is a possibly empty vector if the following is true

sum(size(x) > 1) <= 1 sum(size(x) > 1) > 1 % is a possibly empty vector % is not a possibly empty vector

An array x is a possibly empty row or column vector if the following is true (the two methods are equivalent)

ndims(x) <= 2 & sum(size(x) > 1) <= 1 ndims(x) <= 2 & ( size(x,1) <= 1 | size(x,2) <= 1 )

Add ~isempty(x) & ... for x to be non-empty.

13.8 Matrix

An array x is a possibly empty matrix if the following is true

ndims(x) == 2 ndims(x) > 2 % is a possibly empty matrix % is not a possibly empty matrix

Add ~isempty(x) & ... for x to be non-empty.

13.9 Array slice

An array x is a possibly empty 2-D slice if the following is true

sum(size(x) > 1) <= 2 sum(size(x) > 1) > 2 % is a possibly empty 2-D slice % is not a possibly empty 2-D slice

Chapter 14

Logical operators and comparisons

14.1 List of logical operators

M ATLAB has the following logical operators and & or not ~ xor any all Logical AND Logical OR Logical NOT Logical EXCLUSIVE OR True if any element of vector is nonzero True if all elements of vector are nonzero

14.2 Rules for logical operators

Here is a list of some of the rules that apply to the logical operators in M ATLAB.

~(a & b) = ~a | ~b ~(a | b) = ~a & ~b xor(a,b) = (a | b) & ~(a & b) ~xor(a,b) = ~(a | b) | (a & b) ~all(x) = any(~x) ~any(x) = all(~x)

14.3 Quick tests before slow ones

If several tests are combined with binary logical operators (&, | and xor), make sure to put the fast ones ﬁrst. For instance, to see if the array x is a real positive ﬁnite scalar double integer, one could use

isa(x,’double’) & isreal(x) & ~any(isinf(x(:))) & all(x(:) > 0) & all(x(:) == round(x(:))) & all(size(x) == 1)

but if x is a large array, the above might be very slow since it has to look at each element at least once (the isinf test). The following is faster and requires less typing 44

CHAPTER 14. LOGICAL OPERATORS AND COMPARISONS

isa(x,’double’) & isreal(x) & all(size(x) == 1) ... & ~isinf(x) & x > 0 & x == round(x)

45

Note how the last three tests get simpliﬁed because, since we have put the test for “scalarness” before them, we can safely assume that x is scalar. The last three tests aren’t even performed at all unless x is a scalar.

Chapter 15

Miscellaneous

This section contains things that don’t ﬁt anywhere else.

15.1 Accessing elements on the diagonal

The common way of accessing elements on the diagonal of a matrix is to use the diag function. However, sometimes it is useful to know the linear index values of the diagonal elements. To get the linear index values of the elements on the following diagonals

(1) [ 1 0 0 0 2 0 0 0 3 ] (2) [ 1 0 0 0 0 2 0 0 0 0 3 0 ] (3) [ 1 0 0 0 0 2 0 0 0 0 3 0 ] (4) [ 0 1 0 0 0 0 2 0 0 0 0 3 ] (5) [ 0 1 0 0 0 0 2 0 0 0 0 3 ]

one may use

1 1 1 1 : : : : m+1 m+1 m+1 m+1 : : : : m*m m*n m*m m*min(m,n) % % % % square m-by-m matrix (1) m-by-n matrix where m >= n (2) m-by-n matrix where m <= n (3) any m-by-n matrix

m-n+1 : m+1 : m*n (n-m)*m+1 : m+1 : m*n

% m-by-n matrix where m >= n (4) % m-by-n matrix where m <= n (5)

To get the linear index values of the elements on the following anti-diagonals

(1) [ 0 0 3 0 2 0 1 0 0 ] (2) [ 0 0 0 1 0 0 2 0 0 3 0 0 ] (3) [ 0 0 3 0 0 2 0 0 1 0 0 0 ] (4) [ 0 0 1 0 0 2 0 0 3 0 0 0 ] (5) [ 0 0 0 3 0 0 2 0 0 1 0 0 ]

one may use

m m m m : : : : m-1 m-1 m-1 m-1 : : : : (m-1)*m+1 (m-1)*n+1 (m-1)*m+1 (m-1)*min(m,n)+1 % % % % square m-by-m matrix (1) m-by-n matrix where m >= n (2) m-by-n matrix where m <= n (3) any m-by-n matrix

46

CHAPTER 15. MISCELLANEOUS

m-n+1 : m-1 : m*(n-1)+1 (n-m+1)*m : m-1 : m*(n-1)+1 % m-by-n matrix where m >= n (4) % m-by-n matrix where m <= n (5)

47

15.2 Creating index vector from index limits

Given two vectors lo and hi. How does one create an index vector

idx = [lo(1):hi(1) lo(2):hi(2) ...]

A straightforward for-loop solution is

m = length(lo); % length of input vectors idx = []; % initialize index vector for i = 1:m idx = [ idx lo(i):hi(i) ]; end

which unfortunately requires a lot of memory copying since a new x has to be allocated each time through the loop. A better for-loop solution is one that allocates the required space and then ﬁlls in the elements afterwards. This for-loop solution above may be several times faster than the ﬁrst one

m len n lst = = = = length(lo); hi - lo + 1; sum(len); cumsum(len); % % % % length of input vectors length of each "run" length of index vector last index in each run

idx = zeros(1, n); % initialize index vector for i = 1:m idx(lst(i)-len(i)+1:lst(i)) = lo(i):hi(i); end

Neither of the for-loop solutions above can compete with the the solution below which has no forloops. It uses cumsum rather than the : to do the incrementing in each run and may be many times faster than the for-loop solutions above.

m = length(lo); len = hi - lo + 1; n = sum(len); % length of input vectors % length of each "run" % length of index vector

idx = ones(1, n); % initialize index vector idx(1) = lo(1); len(1) = len(1)+1; idx(cumsum(len(1:end-1))) = lo(2:m) - hi(1:m-1); idx = cumsum(idx);

If fails, however, if lo(i)>hi(i) for any i. Such a case will create an empty vector anyway, so the problem can be solved by a simple pre-processing step which removing the elements for which lo(i)>hi(i)

i = lo <= hi; lo = lo(i); hi = hi(i);

There also exists a one-line solution which is very compact, but not as fast as the no-for-loop solution above

x = eval([’[’ sprintf(’%d:%d,’, [lo ; hi]) ’]’]);

CHAPTER 15. MISCELLANEOUS

48

15.3 Matrix with different incremental runs

Given a vector of positive integers

a = [ 3 2 4 ];

How does one create the matrix where the ith column contains the vector 1:a(i) possibly padded with zeros:

b = [ 1 2 3 0 1 2 0 0 1 2 3 4 ];

One way is to use a for-loop

n = length(a); b = zeros(max(a), n); for k = 1:n t = 1:a(k); b(t,k) = t(:); end

and here is a way to do it without a for-loop

[bb aa] = ndgrid(1:max(a), a); b = bb .* (bb <= aa)

or the more explicit

m = max(a); aa = a(:)’; aa = aa(ones(m, 1),:); bb = (1:m)’; bb = bb(:,ones(length(a), 1)); b = bb .* (bb <= aa);

To do the same, only horizontally, use

[aa bb] = ndgrid(a, 1:max(a)); b = bb .* (bb <= aa)

or

m = max(a); aa = a(:); aa = aa(:,ones(m, 1)); bb = 1:m; bb = bb(ones(length(a), 1),:); b = bb .* (bb <= aa);

15.4 Finding indices

15.4.1 First non-zero element in each column

How does one ﬁnd the index and values of the ﬁrst non-zero element in each column. For instance, given

CHAPTER 15. MISCELLANEOUS

x = [ 0 4 0 0 1 3 0 9 0 7 2 0 0 0 6 5 ];

49

how does one obtain the vectors

i = [ 2 1 2 3 ]; v = [ 4 1 7 6 ]; % row numbers % values

If it is known that all columns have at least one non-zero value

[i, t = i = v = j, v] = find(x); logical(diff([0;j])); i(t); v(t);

If some columns might not have a non-zero value

[it, jt, vt] = find(x); t = logical(diff([0;jt])); i = repmat(NaN, [size(x,2) 1]); v = i; i(jt(t)) = it(t); v(jt(t)) = vt(t);

15.4.2 First non-zero element in each row

How does one ﬁnd the index and values of the ﬁrst non-zero element in each row. For instance, given

x = [ 0 4 0 0 1 3 0 9 0 7 2 0 0 0 6 5 ];

how dows one obtain the vectors

j = [ 1 2 3 1 ]; v = [ 1 4 2 9 ]; % column numbers % values

If it is known that all rows have at least one non-zero value

[i, [i, t = j = v = j, v] = find(x); k] = sort(i); logical(diff([0;i])); j(k(t)); v(k(t));

If some rows might not have a non-zero value

[it, jt, vt] = find(x); [it, k] = sort(it); t = logical(diff([0;it])); j = repmat(NaN, [size(x,1) 1]); v = j; j(it(t)) = jt(k(t)); v(it(t)) = vt(k(t));

CHAPTER 15. MISCELLANEOUS

50

15.4.3 Last non-zero element in each row

How does one ﬁnd the index of the last non-zero element in each row. That is, given

x = [ 0 5 0 8 9 0 0 0 7 0 0 4 0 6 0 2 0 0 0 1 0 3 0 0 ];

how dows one obtain the vector

j = [ 3 6 0 5 ];

One way is of course to use a for-loop

m = size(x, 1); j = zeros(m, 1); for i = 1:m k = find(x(i,:) ~= 0); if length(k) j(i) = k(end); end end

or

m = size(x, 1); j = zeros(m, 1); for i = 1:m k = [ 0 find(x(i,:) ~= 0) ]; j(i) = k(end); end

but one may also use

j = sum(cumsum((x(:,end:-1:1) ~= 0), 2) ~= 0, 2);

To ﬁnd the index of the last non-zero element in each column, use

i = sum(cumsum((x(end:-1:1,:) ~= 0), 1) ~= 0, 1);

15.5 Run-length encoding and decoding

15.5.1 Run-length encoding

Assuming x is a vector

x = [ 4 4 5 5 5 6 7 7 8 8 8 8 ]

and one wants to obtain the two vectors

len = [ 2 3 1 2 4 ]; val = [ 4 5 6 7 8 ]; % run lengths % values

CHAPTER 15. MISCELLANEOUS

one can get the run length vector len by using

len = diff([ 0 find(x(1:end-1) ~= x(2:end)) length(x) ]);

51

and the value vector val by using one of

val = x([ find(x(1:end-1) ~= x(2:end)) length(x) ]); val = x(logical([ x(1:end-1) ~= x(2:end) 1 ]));

which of the two above that is faster depends on the data. For more or less sorted data, the ﬁrst one seems to be faster in most cases. For random data, the second one seems to be faster. These two steps required to get both the run-lengths and values may be combined into

i = [ find(x(1:end-1) ~= x(2:end)) length(x) ]; len = diff([ 0 i ]); val = x(i);

15.5.2 Run-length decoding

Given the run-length vector len and the value vector val, one may create the full vector x by using

i = cumsum(len); j = zeros(1, i(end)); j(i(1:end-1)+1) = 1; j(1) = 1; x = val(cumsum(j)); % length(len) flops % length(len) flops % sum(len) flops

the above method requires approximately 2*length(len)+sum(len) ﬂops. There is a way that only requires approximately length(len)+sum(len) ﬂops, but is slightly slower (not sure why, though).

len(1) = len(1)+1; i = cumsum(len); j = zeros(1, i(end)-1); j(i(1:end-1)) = 1; j(1) = 1; x = val(cumsum(j)); % length(len) flops

% sum(len) flops

This following method requires approximately length(len)+sum(len) ﬂops and only four lines of code, but is slower than the two methods suggested above.

i = cumsum([ 1 len ]); j = zeros(1, i(end)-1); j(i(1:end-1)) = 1; x = val(cumsum(j)); % length(len) flops

% sum(len) flops

15.6 Counting bits

Assume x is an array of non-negative integers. The number of set bits in each element, nsetbits, is

nsetbits = reshape(sum(dec2bin(x)-’0’, 2), size(x));

or

CHAPTER 15. MISCELLANEOUS

bin = dec2bin(x); nsetbits = reshape(sum(bin,2) - ’0’*size(bin,2), size(x));

52

The following solution is slower, but requires less memory than the above so it is able to handle larger arrays

nsetbits = zeros(size(x)); k = find(x); while length(k) nsetbits = nsetbits + bitand(x, 1); x = bitshift(x, -1); k = k(logical(x(k))); end

The total number of set bits, nsetbits, may be computed with

bin = dec2bin(x); nsetbits = sum(bin(:)) - ’0’*prod(size(bin)); nsetbits = 0; k = find(x); while length(k) nsetbits = nsetbits + sum(bitand(x, 1)); x = bitshift(x, -1); k = k(logical(x(k))); end

Glossary

null-operation an operation which has no effect on the operand operand an argument on which an operator is applied singleton dimension a dimension along which the length is zero subscript context an expression used as an array subscript is in a subscript context vectorization taking advantage of the fact that many operators and functions can perform the same operation on several elements in an array without requiring the use of a for-loop

53

Index

matlab faq, 55 comp.soft-sys.matlab, vi, 2, 55 dimensions number of, 7 singleton, 7 trailing singleton, 7 elements number of, 7 emptiness, 8 empty array, see emptiness null-operation, 7 run-length decoding, 51 encoding, 50 shift elements in vectors, 12 singleton dimensions, see dimensions, singleton size, 6 trailing singleton dimensions, see dimensions, trailing singleton Usenet, vi, 2

54

Appendix A

M ATLAB resources

The MathWorks home page

On The MathWorks’ web page one can ﬁnd the complete set of M ATLAB documentaton in addition to technical solutions and lots of other information. http://www.mathworks.com/

The M ATLAB FAQ

For a list of frequently asked questions, with answers, see see Peter Boettcher’s excellent M ATLAB FAQ which is posted to the news group comp.soft-sys.matlab regularely and is also available on the web at http://www.mit.edu/~pwb/cssm/

55