Why find doesn't work for some number?

Hello,

I have the following code

if true
  for i = 0:0.1:2
       k = find (A==i);
       disp(k);
end

This code works but for some number it doesn't work. For example, i=0.6 and 0.7. But when I wrote find(k==0.6) it works!!!

What is the problem? Any idea please

Answer 1

[Adam]

Testing exact equality between double-precision data is un-safe in general. Due to the representation of these numbers a test for absolute equality can fail because of a difference of 0.000001 that you don't see in the variable editor or command window.

If your values such as 0.6 in the array A are the result of some calculation then this will often be the case that they are not precisely 0.6, but may instead be 0.600000000001

[Guillaume]

It is particularly problematic with a step of 0.1 (and plenty of other numbers) as 0.1 cannot be represented exactly in binary.

So your step is actually something like 0.100000000000000005.

[mariem farhat]

Ok thanks for your answers. So, how can I fix this problem?

[Andreas Goser]

In your example: Not comparing with == but a tolerance band you can define yourself based on the application

[Rotem Bachar]

hi,

what does it mean a tolerance band you can define yourself based on the application?

[Rik]

@Rotem Bachar

You can either use ismembertol, uniquetol, etc, or you can do something like this:

% instead of A==B
abs(A-B)<=tol

[Steven Lord]

Close only counts in horseshoes, hand grenades, and comparing floating-point numbers. The tolerance controls what is "close enough" to a match to count as a match.

x = 1;
y = 2;

If you compare x and y with a tolerance of 5 ("close enough" being within 5 units) then x and y match.

doTheyMatch = abs(x-y) <= 5

doTheyMatch = logical
   1

If you compare x and y with a tolerance of 0.5 ("close enough" being within 0.5 units) then they do not.

doTheyMatch = abs(x-y) <= 0.5

doTheyMatch = logical
   0

If you used a tolerance of 0 you're saying "close enough isn't good enough, it has to be an exact match down to the last bit" and in that case sometimes numbers that look like they match don't actually match.

tenth = 0.1; % Not exactly one tenth, but close
z1 = tenth + tenth + tenth

z1 = 0.3000

z2 = 0.3

z2 = 0.3000

doTheyMatch = abs(z1-z2) <= 0 % essentially equivalent to z1 == z2

doTheyMatch = logical
   0

howFarApart = abs(z1-z2) % very small but not 0

howFarApart = 5.5511e-17