Denis P. Mendoza wrote:Eelco, I just edited my profile for ease. Now, you can see my e-mail and site. Your request was also done as well. You can visit your post at http://kirr.homeunix.org/chess/discussi ... 92&p=30403. Multithreading still works when I tested it. Thanks.
Thanks a lot for the compiles Denis! Very good that the multithreading still works! I did not change it directly but I did create some new variables and did not pay much attention if changes to the search could affect the parallel search. I did not expect it but you do not know for sure until you try it out. Also I made for instance a change to the Verification search where I really don't understand why it did not work when I used a negative "Delta Margin", but it did work better with using a postive one. Well it is sort of a optimistic evaluation so it trusts nullmove more, as far as I can see now, and it can help either way, but I had started with a pessimistic eval... I used a constant of 55 but maybe implementing the usual DeltaMargin there will be more elegant. If I just could figure out why exactly a negative one gives worse results
If this is a bad Toga Beta: I will just have to go back to the drawing board but some testing input is welcome, also because testing on a single system by a single method can give biased results as we all know! And in my case I have only played six games with this version!
I just took a first look at the compiles, very good, but I did notice that I may have introduced some last minute changes that affect the engine's behaviour. It is difficult to say what is causing it, I changed some of the default UCI-settings to what I had set in the Shredder GUI, removed the Toga King Safety as UCI option altogether which may have been a bit drastic but at least it was not in use, not by me, and replaced it with Thomas Gaksch' Toga Exchange Bonus which was already in use but not visible anymore as an UCI-option.
Why exactly is a mystery but setting the Toga Exchange Bonus from 35 to 40 seems to give a little better result now just in one or two positions I tested so far. Maybe it is dependent on the computer but I'm hoping that setting the Toga Exchange Bonus to 40 is more or less giving the behaviour that I intended with this version. Please change that if you are willing to play some games.
I leave it up to everybody of course to find settings that may work better in their computers and with their respective set ups, but this change really looks a bit better here, in the short time that I have had the chance to test it.
I will take a look at the sources again if I really changed something other than the options and comments, I suspect it is just a matter of a different compile that produces some changes. But the compiles are faster!
Chris Formula wrote:thanks for raising this issue. there will be an update soon. testing of a good candidate as update version (1.4.2SE) is underway.
Yesterday I took the Space weight evaluation SSpaceWeight[16] from Thomas out of eval.cpp, that term gives a bonus for pieces crossing the middle of the board into enemy territory, I have not had time to do any testing how the playing style works out without it. But it is the only change in eval.cpp itself also from Toga 1.3.1 to 1.3.4 I believe. Of course there are a lot of other things that have changed elsewhere.
