Collision Hash Function

Question

Hi all I have a big problem with my hash function. I try to explain my problem :

I have a set of char and I want to do an hash function because I want to change the set with hash set, for each char I have a index , so what I do now :

pair --> index p = 1 index a = 2 index i = 3 index r= 4---> so my hash return 1234

but if for example I have

so --> index s = 12 index o = 34 ---> hash 1234

COLLISION!!!!

P.S. : I cannot order my char in alphabetic number....

So , is there anyone that can help me?? THANKS A LOT :)

This is why you shouldn't design your own hash. What's wrong with using existing widely-used ones, like md5/sha1? — Marc B, Apr 17 '13 at 15:21
Commonly the chars are multiplied with some good chosed numbers which often gets XORed, but you do nothing like that... Typical multiple mathematical experts needs month to develope a hash algorithm. — rekire, Apr 17 '13 at 15:22
Because I have also integers numbers in my set and if I have to do to_string and then pass the value to md5/sha1 is too much expensive :( — Leonardo Rania, Apr 17 '13 at 15:25
Building a hash is more expensive than converting. Depending on your data structure you could simply hash the memory of your structure. So you don't need to convert anything. — rekire, Apr 17 '13 at 15:30
No no trust me...building a hash is more expensive only for the brain :) I tried to use md5/sha1/superFast... — Leonardo Rania, Apr 17 '13 at 15:33
Try a bit rotation together with XOR the next byte. That would result less collisions, but that will produce some. By the way if you don't be an expert or a PhD i won't trust you ;-) — rekire, Apr 17 '13 at 15:37
Yes,I didn't say that is always expensive but in my case yes :) however I'm doing a master thesis... — Leonardo Rania, Apr 17 '13 at 15:44
In this case I'm just a little step ahead I've completed mine. — rekire, Apr 17 '13 at 15:48
xor is not good for me ..... my hash function now is : h = join(h,CHAR.getIndex()+1) — Leonardo Rania, Apr 17 '13 at 15:52
Which programming language you using? XOR should be fast AFIK XOR is also a CPU command. — rekire, Apr 17 '13 at 15:58
c++ .... sure xor is fast but not in my case I have to found a number to sum or multiply or other operation to CHAR.getIndex() to resolve the collisions ....but my hash function must be h = join(h,CHAR.getIndex()).... — Leonardo Rania, Apr 17 '13 at 16:02

rekire · Answer 1 · 2013-04-18T03:56:44.223

0

You could try the string hashing function of Java. This is my C# port which should be simple ported to c++:

int javaHash(String txt) {
    uint h = 0;
    if(txt.Length > 0) {
        for(int i = 0; i < txt.Length; i++) {
            h = 31 * h + txt[i];
        }
    }
    return (int)h;
}

edited Apr 18 '13 at 03:56

answered Apr 17 '13 at 16:30

rekire

47,260
30
167
264

Collision Hash Function

1 Answers1