Author: Melvin S. Schwartz
Date: 22:01:40 07/26/99
Go up one level in this thread
On July 25, 1999 at 14:48:06, Mark Young wrote: >On July 25, 1999 at 10:59:28, Melvin S. Schwartz wrote: > >> >>On July 24, 1999 at 23:27:06, Mark Young wrote: >> >>>On July 24, 1999 at 22:56:28, Melvin S. Schwartz wrote: >>> >>>> >>>>On July 24, 1999 at 21:18:38, Mark Young wrote: >>>> >>>>>On July 24, 1999 at 19:22:26, blass uri wrote: >>>>> >>>>>>I do not know enough about chess to decide if the moves a4 and h4 are the right >>>>>>moves. >>>>>> >>>>>>It is clear that my hiarcs evaluated a4 as not the best move after 3 hours and >>>>>>prefered Rf1 but it only proved that my hiarcs believed after a long time that >>>>>>a4 is not the best move(Hiarcs7 may be wrong in this evaluation so it does not >>>>>>prove that a4 is not the best move). >>>>>>I do not know enough about chess to know which move is really best in the >>>>>>position. >>>>>> >>>>>>I do not see a problem with the fact that another hiarcs has another opinion >>>>>>because of the following possible reasons: >>>>>>1)another hiarcs may have another learning file >>>>>>2)Hiarcs is not deterministic >>>>> >>>>>You are correct Hiarcs is not deterministic, as you know my hiarcs 7.32 sticks >>>>>with a4 after 3 hours and depth 12. But the issue never was would hiarcs 7.32 >>>>>play another move. That issue only came about because this is what Mel based his >>>>>opinion on that a4 must be a blunder because other >>>>____________________________ >>>> >>>>I do not consider what I say is a bad move to mean the same thing as a blunder. >>>>A blunder changes the position dramatically. A bad move is simply a poor choice >>>>when there is a better move. It is not because my other programs felt a4 was bad >>>>that I said what I did. I stated a4 was a bad move because Hiarcs refused to >>>>play it again and insisted on Rf1 with a better score. In my opinion, if a >>>>program selects one of two moves that results in a clearly better score, then >>>>the other move must be considered a bad move for the program would not repeat >>>>it. I believe that is quite logical thinking. >>>>___________________ >>>> >>>>programs don't play it and >>>>>because hiarcs is not deterministic it choose another move when mel forced it to >>>>>think on the position again. Then came to the mistaken conclusion hiarcs 7.32 >>>>>did not play a4 again because its leaning fuction saw it as a mistake. Then he >>>>>takes this mistaken canclusion about this position as proof that Hiarcs 7.32 is >>>>>not that good. When the facts are he never looked at the position himself and >>>>>has no idea if a4 is good or bad in fact. Very reckless..... >>>>____________________ >>>> >>>>There was no need for me to look into why Hiarcs would not repeat a4 when it is >>>>quite apparent from the better score it gave with Rf1 was evidently a function >>>>of the learning process - a logical conclusion based on what the program >>>>evaluated and insisted on playing. If it wasn't the learning function that >>>>prohibited a4 from being repeated, then I must assume Hiarcs is one dizzy >>>>program. :) >>>>_______________________________> >>> >>>Its not the program that is dizzy. :) >>____________________________ >> >>Well, it certainly isn't me. I cannot understand how you can refute such a >>logical conclusion that Hiarcs learning function was at work here. When a >>program makes a move and its score drops and then you let it replay the move and >>it selects a different move with a better score, what do you think was the >>reason that program will not repeat the first move but only the second move >>which resulted in a better score? Hmmm? What is the learning function for? I >>should also state that I have experimented with this before - letting Hiarcs >>replay a move when its score dropped - of the few times I've done that, this is >>the first time it actually changed its move, though I should add this was the >>biggest drop when I let it replay the move.This a4 thing may not look bad to you >>but apparently Hiarcs saw something when the score suddenly went down to a minus >>and it HAS to be the learning function that would not let Hiarcs repeat a4. If >>you cannot reproduce the same situation on your program as what happened in my >>game, that does not mean I am dizzy. The fact you were not actually playing a >>game might have something to do with it if your Hiarcs does not go from a plus >>to a minus AFTER making the move a4. Whatever the reason, what occurred as I >>have stated is reality, and not a story from the "Twilight Zone". > >To be clear it don't doubt Hiarcs 7.32 played other move, but this had nothing >to do with the learning fuction. Since you never played a game against hiarcs >7.32 with the move a4 because you took the move back, ************** Mark, I didn't take the move back, I simply let Hiarcs replay the move. I didn't think a4 was a good move based on my opinion of the position, and when the score went way down while I was thinking, I couldn't help but let Hiarcs replay the move to see if I was right and Hiarcs saw it was a poor choice. Now, I let Hiarcs replay the move TWICE, and each time it selected Rf1 with a much better score. Therefore, the only logical conclusion that can be drawn from this is that the learning function would not permit a4 because it resulted in a minus score and Rf1 was a plus score of around a half-pawn better. That indicated to me Hiarcs felt Rf1 was definitely a better move. That is what the learning function is about - the program will not repeat a move it finds to be bad and looks for a better move. Hiarcs 7.32 could not >learn, as it does not know if the move wins, or loses or draw, or drops >material. ********** Mark, Obviously Hiarcs saw something after it played a4 that caused the score to drop about a half-pawn. What is so difficult to believe that it wouldn't play a4 again because Hiarcs saw it was a bad choice after making the move? There is no other logical explanation for the reason it would not repeat a4 but chose to keep playing Rf1 because it found a better line with a better score of around a half-pawn which is quite substantial. The only time Hiarcs will change a move like that with such a score difference is because the learning function was at work and would not repeat a4. That is my opinion based I believe on a valid result. If you have a better explanation as to why Hiarcs would not repeat a4, I would like to hear it. Mel> > >*2)Hiarcs is not deterministic*: Mel do you know what Uri is talking about when >he said Hiarcs is not deterministic? That is the most likely answer to why it >changed for you, and did not change for me, but could have. > >The last thing I will as on this subject is, it is a mistake for you to just >site the difference in a programs eval score as proof a move is weaker or a >blunder without ever looking at the position for yourself. *********** I did not say it was a blunder. I said it was a bad move which means a bad choice. As I've said before, a blunder changes a position dramatically where a bad move is simply a poor choice. As for Uri, I believe he assumed my Hiarcs did not play a4 but played Rf1. I don't think he was aware of how my Hiarcs also originally played a4 juat like yours. It is not just the evaluation score that made me say it was a poor choice - it was that coupled with the fact it would not repeat that move! We you say move(a) >has a score of +.34 and move(b) has a score of -.34 then site this as proof that >move(b) is a blunder and you use only a programs eval as proof, it is very hard >to take your criticisms of Hiarcs 7.32 move choice seriously. ******** You are completely ignoring the fact Hiarcs would not repeat the move. The only logical reason a program will not repeat a move - when it realizes after making the move it evaluates it as about a half-pawn less then what it had before making the move - is because it found a better move. I think it is as simple as that. >>> >>>>>> >>>>>> >>>>>>About the second position: >>>>>>7k/6p1/7p/pR6/1p2N3/6P1/2r4P/6K1 w - - 0 1 >>>>>> >>>>>>I think that h4 is leading to a draw and the same for Nc5 >>>>>>It is easy for white to get draw after Nc5 Ra2 (for example Nb7 a4 Rxb4 a3 >>>>>>Ra4 Rb2 Rxa3 Rxb7) >>>>>> >>>>>>white do not have to play Nb7 but I do not see a way for white to make progress >>>>>>in a different way. ******************* My response to Rc2 is Kf1. I believe White can get a draw if needed but has winning chances by playing Nc5. I don't see any reason not to take the gamble with Nc5 which should at least draw if not win.If after Kf1 you take the h-pawn, I will take the a-pawn. Therefore, Black cannot take the h-pawn but I will advance my h-pawn and...all I can say is Black does not in my opinion have winning chances. One of those passed pawns will soon disappear, and the game will be left with Black having one passed pawn that will not be a Queen in my opinion. Why don't you play Hiarcs as Black against Rebel 10 as White and see what you get? That should be interesting since Hiarcs 7.32 is supposed to be better than Rebel 10. Of course we are not playing Hiarcs against me, but I think Rebel will do just fine. I am talking 40/2 here and not blitz. Check it out and see what develops. >>>>>> >>>>>>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.