Author: Ricardo Gibert
Date: 21:19:19 05/08/05
Go up one level in this thread
On May 08, 2005 at 05:17:25, Uri Blass wrote: >On May 08, 2005 at 03:35:49, Ricardo Gibert wrote: > >>On May 07, 2005 at 09:14:27, Uri Blass wrote: >> >>>On May 07, 2005 at 05:46:32, Claude Le Page wrote: >>> >>>>I suggest another method: >>>>consider the famous game Atars-Tomson (it is in every database) >>>>1 e4 e5 2 Nf3 f5 3 Bc4 fxe4 4 Nxe5 Qg5 5 d4 Qxg2 6 Qh5+ g6 7 Bf7+ Kd8 >>>>8 Bxg6 Qxh1+ 9 Ke2 c6 10 Nc3 Kc7 11 Bf4 Qxa1 12 Nxd7+!! (every "good" engine >>>>finds it)Kxd7 13 Qf5+ Kd8 14 Qxf8+ Kd7 15 Bf5# >>>>The interesting point is elsewhere:On 12...Kd8 (best!)there is #14 (generally >>>>announced as #18 ) >>>>There are at least 3 mates completely different: that of Junior7 and that of >>>>Shredder9 have nothing in common (although both are #14 announced as#18) >>>>It would be interesting to compare on his position PVs of Patriot , Toga and so >>>>on >>>>As I am not Programmer , Just Analyst ( in my previous job as in my chess life ) >>>>this method is easier for me >>>>Surely , the position I quoted is not unique , and you could easily find other >>>>ones >>> >>>What you suggest is not another method. >>> >>>Of course analyzing positions can practically help to prove that 2 engines are >>>clones and I think that the test to detect clones should include also secret >>>positions to see if the move of the program is the same as another open source >>>program. >>> >>>I think that it is not productive to talk here about the exact methods to >>>discover clones because it can help cloners and it is better to talk about it by >>>email. >>> >>>Uri >> >>How about a dozen test positions where the position is blocked and one move is >>as good as another. Only a clone would show the same PV in all or most of the >>test positions. > >Of course you can use them but I do not think that they are enough(see later). > >> >>Such positions would be more effective than the usual test positions that often >>have just one line that is definitely best. > >I did not suggest to use positions from known test suites and of course >positions when there are many good moves can be productive to detect clones. > >The idea is to generate set of positions when every 2 engines show different >output in them when clones shows similiar output. > >The problem with your test is that clones may show also different output and you >may get different output if you do small changes in the piece square table of >fruit. I worried a little about this. Certainly the idea is probably too sensitive to small simple changes to the clone. Still, it might be useful in showing that 2 different engines are clones without danger of a false positive for "cloneliness". It's no magic bullet against clones, but perhaps a little bit useful nonetheless. Other types of test positions have too great a danger of producing a false positive. How to avoid this? > >Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.