Re: Spellcheck.adb

[Albert Bachmann <albert.bachmann@gmx.de>]

On Thu, 28 Apr 2005 00:58:29 +0000, Matthew Heaney wrote:

> Albert Bachmann <albert.bachmann@gmx.de> writes:
> 
>> Yes. That's it. I should have known. And timing is not bad with hashed_set:
>> 
>> -- with custom hash function:
> 
> What is your custom hash algorithm?
> 

Just the one used in the C++ entry. With "custom" I just wanted to
distinguish it from Ada.Strings.Hash.

with ada.containers;
use  ada.containers;

function hash_function(key : string) return hash_type is
	n : hash_type := 0;
begin
	for i in key'range loop
		n := (n * 5 + character'pos(key(i)));
	end loop;

	return n;
end hash_function;

> 
>> -- with ada.strings.hash:
> 
> I just checked in Pascal Obry's improved algorithm for ada.strings.hash.
> See if it does any better on your input.
> 
> 
>> Using streams makes for another 10ms. But the GNAT.Spitbol.Table still
>> seems to be the fastest version so far.
> 
> What were those numbers again?
> 
> Regards,
> Matt

With David Sauvage's version (with uses GNAT.Spitbol.Table) I get the
following times:

real    0m0.074s
user    0m0.054s
sys     0m0.012s

I have no specific times for the streams version at hand since I deleted
it already. But I remember that back then the difference was in the range
of 10ms. Note that David Sauvage's version uses streams which contribute
to his excellent results.

http://charles.tigris.org/source/browse/charles/src/ai302/examples/spellcheck/
still shows only this initial revisions and the hash function there is the
same as the one I used.

Regards,
Albert