result of ismember is wrong in my code correct in command window

Dear, I have this code, when I get the result from the loop I find the common rows between pp and pointsout but I got wrong result, when I copy my matrix pp and pointsout to command window or I comment my code and tried from file it gives correct answer. for example

pp =[
   29.6360   20.2328   25.8546
   22.0039   27.8829   29.8251
   23.3343   26.8422   20.0174
   20.1460   29.4970   29.2674
   27.7054   29.9161   24.0945
   29.9539   27.4832   29.3899];
pout=[];
pin=[];
points=pp;
for k=1:size(pp,1)
[ClosestPoint,Distance] = Cl(XYZ,pp(k,:));
if Distance ==-2 
 pin=[pin; pp(k,:)]
end
pout=[pout; ClosestPoint];
end
pointsout=[pointsin; pout];
% find the index of the in points
idx=ismember(pointsout ,points,'rows')

if

pointsout =[
   33.6612   39.8834   33.3147
   25.2505   41.9976   32.2839
   25.6122   41.9067   32.3282
   23.0877   42.5413   32.0188
   29.4130   40.9513   32.7941
   32.6676   40.1332   33.1929];

it gives me just five rows by ismemebr, when I take the two matrices and check in the command window I get 6 rows even when I comment the code to check. What is wrong of the code, why it effect the result. I'm sorry I can't write the whole code for respect to function Cl its 4 files. Thanks for any comments and help.

Nadia,

Answer 1

[James Tursa]

I suspect this is just floating point issues. Reading from a file may read in all of the precision of the numbers (i.e., precision beyond the 4 digits after the decimal point you are showing). But setting the numbers directly from the command line as in your code above may give you slightly different results. I suggest you check and compare the pp and pointsout etc you get from the various methods you are using to see if there are any differences.

[nadia nadi]

Dear James,

thanks for replying quickly.

there is no different in the decimal points, its the same. I suspect in my code, do think the way that I wrote the loop and the result may effect because as I said when I comment the code in my fiel and apply ismember for the matrices its ok.

regards

[Stephen23]

Nadia Nadi: it is very likely that it is floating-point issues, exactly as James Tursa replied. You should download this:

http://de.mathworks.com/matlabcentral/fileexchange/22239-num2strexact--exact-version-of-num2str-

and compare the values that you think are identical. Most likely they are not. You will learn that what is printed in the command window is not the same as the value stored in memory, it is only a very simplified version of of the value.

[nadia nadi]

Dear Stephen,

I think that's correct look at this example, these matrices give me one similar row in the command and from my file, I used format long and short and both of them give me incorrect result. If I want to find ismember just for 4 digits 26.0140 what should I do ?, because format can not help. I'm searching for some ideas know, I wrote single (PP), but it gives the same so is better to use 4 digits, so I used

PP=round(PP,5)

its the same, any idea that you can help me please? regards,

Nadia

[Guillaume]

You have two options to fix the problem:

1. use ismembertol introduced in r2015a. I suspect that ismembertol is significantly slower than standard ismember but for small sets, it probably doesn't matter
2. round the values before passing them to ismember. Should be faster (and simpler) than option 1:

idx = ismember(round(pointsout, 4) ,round(points, 4),'rows')

[nadia nadi]

Dear Guillaume,

thanks that is great, this works well in r2015a, but not in r2012b, because round(points, 4) does not work in it. Do you know a way to use it in r2012b. that will be better because all my files run from it.

regards

[Star Strider]

You can emulate the most recent round function with this one:

roundn = @(x,n) round(x .* 10.^n)./10.^n;   % Round ‘x’ To ‘n’ Digits

[nadia nadi]

Thank you very much, it was all good advice. I appreciate all of you.