MATLAB Central Newsreader - problem with the find function in matlab

Re: problem with the find function in matlab

Steven Lord — Mon, 07 Jan 2008 00:43:51 -0500

"koolguyuf" <tdgoswami@gmail.com> wrote in message
news:bf28c7fc-c160-40da-ac3b-0e7b48b025b1@v46g2000hsv.googlegroups.com...
> Hi,
>
> I am facing some weird behavior using the find function in matlab
> versions 7.0 and 7.5 . I have this extremely simple code to execute.
>
> a = [0:0.2:2];
> my_index = find(a == 0.6)
>
> plz let me know whether you are getting my_index = 4 ? Suprisingly, I
> am not able to get it.

This is a very frequently asked question.

http://matlabwiki.mathworks.com/MATLAB_FAQ

Question 6.1 talks about this type of behavior.

--
Steve Lord
slord@mathworks.com

Re: problem with the find function in matlab

Roger Stafford — Sun, 06 Jan 2008 03:46:36 -0500

"Jerecho " <jeric3182@mathworks.com> wrote in message <flpgo6$ouf
$1@fred.mathworks.com>...
> Im also experiencing the same problem. But when I
> entered 'a' manually (i.e. a = [0 0.1 0.2 0.4 0.6 ... 2])
> I get the correct answer.
-------
That should tell you what the trouble is, Jerecho. It's not the fault of 'find'. It's
because 0:.2:2 generates a slightly different vector of values than [0,.2,.4,.6,
etc]. And that in turn is because it does not have an exact value of .2 to work
with. It cannot achieve .2 precisely and any calculations with it are bound to go
slightly wrong at some stage.

Roger Stafford

Re: problem with the find function in matlab

Jerecho — Sun, 06 Jan 2008 03:08:54 -0500

Re: problem with the find function in matlab

Roger Stafford — Sat, 05 Jan 2008 20:52:05 -0500

koolguyuf <tdgoswami@gmail.com> wrote in message <bf28c7fc-
c160-40da-ac3b-0e7b48b025b1@v46g2000hsv.googlegroups.com>...
> Hi,
>
> I am facing some weird behavior using the find function in matlab
> versions 7.0 and 7.5 . I have this extremely simple code to execute.
>
> a = [0:0.2:2];
> my_index = find(a == 0.6)
>
> plz let me know whether you are getting my_index = 4 ? Suprisingly, I
> am not able to get it.
>
> Plz help.
>
> Thanks
> TD
---------
The problem is not with the 'find' command but with the floating point
representation of 0.2 in a. The n-th term in a vector expression, [p:d:q],
should ideally be p+(n-1)*d. However, the difficulty is that this latter
computation may experience round-off errors due to the imprecision of the
computer's representation of such a d. After all, it is using binary not
decimal floating point numbers. Using format hex with your example we get

0+(4-1)*.2 = 3fe3333333333334

whereas the most accurate representation of .6 is actually

.6 = 3fe3333333333333

which is just slightly different in the least bit (as Shaun has also pointed out.)
When you write find(a==.6), the find command requires exact equality of
both sides of the equation and would therefore not pick up the fourth index.

MathWorks uses various schemes in computing progressions of the form
[p:d:q] to avoid some of these eccentric results, but it is unlikely they are able
to avoid all of them. If you have a value of d which has an exact expression
in decimal but not in binary, it is likely that a long enough sequence will
encounter cases that would be missed by a 'find' scan, even in their latest
matlab versions.

Roger Stafford

Re: problem with the find function in matlab

dpb — Sat, 05 Jan 2008 16:23:02 -0500

koolguyuf wrote:
> Hi,
>
> I am facing some weird behavior using the find function in matlab
> versions 7.0 and 7.5 . I have this extremely simple code to execute.
>
> a = [0:0.2:2];
> my_index = find(a == 0.6)
>
> plz let me know whether you are getting my_index = 4 ? Suprisingly, I
> am not able to get it.

W/ R12 it returns 4.

What does it return for you?

What platform?

If not 4, one would have to presume there's an issue of exact matching
for the particular combination of OS/version/hardware.

To get to the bottom of behavior, need above information at least.
Executing the lines and pasting results could be useful as could

a = [0:0.2:2];
b = a - 0.6

just possibly be educational...

--

Re: problem with the find function in matlab

Huy — Sat, 05 Jan 2008 16:11:12 -0500

Re: problem with the find function in matlab

Shaun — Sat, 05 Jan 2008 16:05:59 -0500

>> format hex
>> a = [0:0.2:2];
>> a(4)
ans =
3fe3333333333334
>> 0.6
ans =
3fe3333333333333

Look into floating point numbers. There are several posts

Re: problem with the find function in matlab

Huy — Sat, 05 Jan 2008 16:05:18 -0500

problem with the find function in matlab

koolguyuf — Sat, 05 Jan 2008 15:54:45 -0500

Hi,

I am facing some weird behavior using the find function in matlab
versions 7.0 and 7.5 . I have this extremely simple code to execute.

a = [0:0.2:2];
my_index = find(a == 0.6)

plz let me know whether you are getting my_index = 4 ? Suprisingly, I
am not able to get it.

Plz help.

Thanks
TD