http://qs321.pair.com?node_id=755386


in reply to Re^2: False positive on inequality comparison
in thread False positive on inequality comparison

Can I suggest that you dump each value to a file along with some natural identifier (line number or cell ID), using: 

print unpack 'b64', pack 'd', $float;;
1011110110001011101100101110110101100111111000110111101000000011

[download]

Do this immediately after reading it into your script to ensure the minimum possibility of 'internal influences'. Then, when you encounter unexpected/unexplainable differences, dump both values in the same form, along with their original identifiers.

I can't offer an explanation for how or why what you are seeing could happen, but I have seen some weirdness in Perl's internal handling of floats in the past. Dumping bit patterns is surest way I know of inspecting what is actually there.

That said, you can still be fooled, as IEEE allow for the possibility of denormalised representations. That is, the bit patterns can be different but represent (exactly) the same values. Kind of like binary equivalent of 10e-6 vs. 1e-5.

Anyway, if you can track what differences are being perceived, then you may be able to backtrack to where they arise. And if you can do that, an explanation of why may be obvious or forthcoming.

Good luck.

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
"Science is about questioning the status quo. Questioning authority".
In the absence of evidence, opinion is indistinguishable from prejudice.

Replies are listed 'Best First'.
Re^4: False positive on inequality comparison
by Nkuvu (Priest) on Apr 06, 2009 at 18:08 UTC

    For the record, I did try this as well, but it didn't help my Monday morning mind narrow anything down. Specifically, I found the same three differences. The binary representations didn't change between the time I read the values in and the time I compared them.

    First two lines are reading files in from Excel sheets, second two are later in the script when the difference is found. I didn't include all three differences here, since they're all the same scenario. Different values read in from the files, no change to those values for the comparison:

    read var_f in from change file: 10011100010100111000101110010000100001
+01000001010110011000000010
read var_f in from range file : 10101100010100111000101110010000100001
+01000001010110011000000010
var_f value in change hash    : 10011100010100111000101110010000100001
+01000001010110011000000010
var_f value in range hash     : 10101100010100111000101110010000100001
+01000001010110011000000010

    [download]

    I'm not familiar enough with floating point representation to say what the difference is (it's been years since I had to convert a binary to float, and unfortunately need to do actual work this morning rather than brush up on the process). I think this is a difference in exponents, but that wouldn't make a lot of sense to me with the rest of the binary value being identical.

[reply]
[d/l]

      Hm. When I unpack those two values, they are distinctly different: 

      printf "%17.17f\n", unpack 'd', pack 'b64', 
'1001110001010011100010111001000010000101000001010110011000000010';;
181.01965800261112000

printf "%17.17f\n", unpack 'd', pack 'b64', 
'1010110001010011100010111001000010000101000001010110011000000010';;
181.01965800261101000

$first = unpack 'd', pack 'b64', 
'1001110001010011100010111001000010000101000001010110011000000010';;

$second = unpack 'd', pack 'b64', 
'1010110001010011100010111001000010000101000001010110011000000010';;

printf "%f %f ;> %17.17f\n", $first, $second, $first - $second;;
181.019658 181.019658 ;> 0.00000000000011369

      [download]
      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]

In Section Seekers of Perl Wisdom