Recently some folks at The MathWorks wondered about the difference of behavior of variable creation versus assigning values into an existing array. I thought I'd address this topic live from The MathWorks Aerospace and Defense Conference '07.
Contents
Here's a picture of the audience.
The Question
Why doesn’t changing an element of a logical array change its logicalness? Most arithmetic operations remove the logicalness from an array; it would seem to me that changing an element of the array (at least to something other than 0 or 1) should either throw an error or similarly modify its logicalness.
An Example
Let's show an example. First create a logical array.
clear all
A = magic(3);
B = A > 6;
whosName Size Bytes Class Attributes A 3x3 72 double B 3x3 9 logical
Now let's do some arithmetic with logical arrays mixed with doubles.
C = B + 1; D = B; D(1:5) = 2 whos A B C D
D =
1 1 0
1 1 1
1 1 0
Name Size Bytes Class Attributes
A 3x3 72 double
B 3x3 9 logical
C 3x3 72 double
D 3x3 9 logical
What you see is that A and C are double arrays, while B and D are logical.
The Answer
Indexed assignment, that is, D(1:5) = 2 has always preserved the type of the array being indexed. Only values may change, not the type. You'd only get an error if there was not a valid conversion routine from the type on the right-hand side to the type stored on the left.
If you simply concatenate values, you also find that the output is logical:
E = [D ; B]
E =
1 1 0
1 1 1
1 1 0
1 0 0
0 0 1
0 1 0
To see this with another datatype, let's look at string array.
firstname = 'Loren'
lastname = [83 104 117 114 101]
name = firstname
name(end+2:end+1+length(lastname)) = lastnamefirstname =
Loren
lastname =
83 104 117 114 101
name =
Loren
name =
Loren Shure
If I just use straight concatenation, here's what happens
newname = [firstname, ' ', lastname]newname = Loren Shure
Arithmetic
If I perform arithmetic between mixed types, where one is numeric and one is not, I instead get a numeric answer.
name - 32
ans =
Columns 1 through 8
44 79 82 69 78 -32 51 72
Columns 9 through 11
85 82 69
Thoughts?
The fact that you get results with different datatypes depends on whether you are doing a numerical calculation or assigning to an array. Has the behavior been what you need or have you found it confusing? Let me know here.
Get
the MATLAB code
Published with MATLAB® 7.4
I remember this being very confusing back in R12 when double arrays, the full 64 bits, were tagged with a “logical” flag which would mean, zero is false and, non-zero was true. Very confusing. Thanks for this explanation!
I just did the following experiments:
>> clear
>> x = logical([1 0 1 0 0])
x =
1 0 1 0 0
>> whos
Name Size Bytes Class Attributes
x 1×5 5 logical
>> x(1) = 2
x =
1 0 1 0 0
>> whos
Name Size Bytes Class Attributes
x 1×5 5 logical
I think Matlab should complain when I run x(1) = 2. Matlab does not do the operation I ask it to do and it does not complain. I think this is bad. Instead Matlab should generate error message and let the programmer to fix the problem.
How does cat choose the ‘dominant’ class, i.e., the class of its return value?
A quick experiment with the basic types seems to reveal a descending priority of:
char
integer types (if there are multiple different integer-typed operands, the type of the left-most of these is dominant)
single
double
logical
Appropriate warnings are given most of the time, but truncation of floats to char doesn’t throw any. And promotion of one integer type (or logicals) to a bigger integer type DOES throw a warning, even though it’s unnecessary.
Is there a particular reasoning behind all this behaviour? I haven’t noticed it documented anywhere.
Any non-zero real element of input array X is converted to a logical 1 while zeros in X become logical 0.
which makes a lot of sense, by the way…
us
SNIP misunderstanding
x=[false,true,false];
x(1)=2;
% x = 1 1 0
I think Matlab should complain when I run x(1) = 2…
not at all; the help clearly says:
…
Any non-zero real element of input array X is converted to a logical 1 while zeros in X become logical 0.
…
which makes a lot of sense, by the way…
us
sorry, first reply got screwed up
Wes,
There’s a page in the MATLAB documentation that discusses concatenation of different data types:
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/matlab_prog/f1-84864.html
This is from the documentation for MATLAB 7.4 (R2007a) but I don’t believe this behavior has changed recently. The priority list you posted agrees with what’s in the documentation. This page also discusses the integer casting warning.
I would have to join the group saying Matlab should complain a bit more when it comes to type conversions. Every now and then I have numbers in integer types and I want to keep them that way. However, Matlab tends to like to make them doubles every now and then and tracking down where it happens can be a pain.
However, in the philosophy that Matlab should be math and not programming, I would have to say that warnings are sufficient. Calling it an error would be way overkill.
we want to more example.
Hai guys,
i am currently doing final year project using vision system. my project have to differentiate plastic and paper. I am using normal web-cam to capture the images.The programme that i am using is matlab version 7.0. I am now searching on how to differentiate it. Is there any commands in matlab that can help me in differentiating this two materials. I really want to know if there is any. Please if anyone know do help me.
Thank you
is there any command to differentiate plastic and paper using the image captured by matlab 7.0 and normal web-camera…plz help
jai: How about you take a look at the Matlab file-exchange. There are loads of files there that do all sorts of things. If someone else has wanted to do that particular thing, there will probably be a script.
On the other hand, maybe you should consider doing your own homework. This sounds very much like you are planning to plagiarize someone else’s work and hand it is as your own.
–DA
Hi, I want to know how to perform the arithmetic operation for two image? I have a problem for two image with different size.I really want to know is anyone has a sample coding for that?Plz…help me to find this kind of solution…tq