Author: Christophe Theron
Date: 21:24:00 11/16/00
Go up one level in this thread
On November 15, 2000 at 07:28:57, Gordon Rattray wrote:
>On November 14, 2000 at 13:24:45, Christophe Theron wrote:
>
>[snip]
>
>>GT should handle underpromotions. At least if you give the program enough time
>>it should find the key underpromotion move.
>>
>>One thing to know, maybe it's not what happened for you, but I tell it anyway:
>>the Auto232 protocol is lousy about underpromotions. With the standard Auto232
>>protocol, games will just stop or incorrect moves will be played when an
>>underpromotion occurs. To fix this, you can set the "extended mode" in the
>>Autoplayer dialog. When this box is checked, Tiger will try to transmit
>>underpromotion using an extension of the standard protocol. If the program at
>>the other end of the line knows this extended protocol, then underpromotions
>>will not stop the games anymore.
>>
>>Can you post the offending position, please?
>
>
>My test position was the following:
>
>rk6/p2p4/KPRp4/8/3P4/8/8/8/ w - -
>
>Solution: 1) Rc8+ Kxc8 2) b7+ Kb8 3) d5 Kc7 4) bxa8/B Kb8 5) Bb7 and white
>wins
>
>GT doesn't see this. Maybe the stalemate aspect is playing a role, because when
>I now try other positions involving underpromotions, GT does see it from far
>off. e.g. I tried the case with promotion to a knight in order to fork king and
>queen. GT sees this immediately from multiple moves away.
You are right about this problem. I don't understand where it comes from, and I
will investigate on this.
>>>I've seen a position with one forced move but GT still spent time considering
>>>its reply?!
>>
>>
>>There are two cases:
>>1) if there is only one legal move, GT will play instantly.
>>2) if there is an obvious move that looks forced, GT will play much faster than
>>usual, but it will still spend a fraction of the usual time to check if there is
>>a trap behind this "forced" move. Sometimes it saves the day by doing so.
>
>
>Ok, I have to take this one back as I cannot reproduce it. Initially, I
>basically had a king on a1 being checked by an undefended queen on b2. GT was
>considering its reply so I looked to see what else could capture the queen, and
>I didn't see any other captures. Anyway, I'll have to accept I was
>hallucinating until I can reproduce it. ;-)
>
>
>>>I've arrived at a stalemate position without being told it was stalemate - it
>>>just stopped?!
>>
>>
>>Can you post the position?
>
>
>Upon further investigation, I notice that this happens if and only if I
>stalemate GT and not the other way around. It applies it checkmate too
>(unfortunately I've haven't succeeded in checkmating GT in a normal game, only
>from a test postion ;-) ). e.g. try Black king on e4, Black pawn e3, White king
>e1. Play as Black against GT and advance the pawn to a stalemate position. I
>think the "Statistics" and moves windows give some indication that the game is
>over, but the normal pop up window is not shown. Incidentally, if GT gives the
>stalemate, it says "Stealmate" and not "Stalemate". ;-)
I hope Lex reads this.
>>>Switching analysis mode on/off can be flaky. I can sometimes toggle this button
>>>without seeing any analysis. I often have to use this in conjunction with "make
>>>computer move". Then, depending on the time control, the computer will move,
>>>despite being in analysis mode.
>>
>>
>>I have never noticed this. If you know how to reproduce it, then Lex will be
>>interested.
>
>
>Ok, actual steps to reproduce...
>
>i) start with a new game
>ii) press analysis button on toolbar -> analysis starts to appear in
>"Statistics" window
>iii) edit position, and then come back out of edit mode
>
>Now I see no further analysis produced. Toggling the analysis button doesn't do
>anything.
Again I hope that Lex will look at this.
>>>Sometimes when trying to move a piece with the mouse, if you "miss" the piece,
>>>the whole board window moves. This can be annoying.
>>
>>
>>That's a feature. You can move the board by just clicking on it and dragging it
>>around. If you are annoyed by this, you can create another layout in which the
>>chessboard will have a "window border" (Extra->Options->ChessBoard). When the
>>board has a "window border", you will not be able to drag it by clicking on it.
>>You will be able to drag it by clicking on its title bar, like any other window,
>>and it will solve your problem.
>>
>>Maybe you can just choose another layout from the list of available one and find
>>one that suits you.
>
>
>I looked at the properties for my current board window and it does have a border
>(I can also see that it has! ;-) ). But it is still possible to drag the board
>by clicking on it?! Not a huge issue.
>
>
>>>I don't like the method of setting levels. I prefer a dialog with options and
>>>values.
>>>
>>>I can't see anyway of displaying the time for the current move. I only see the
>>>accumulated time.
>>
>>
>>You can check Extra->Options->View->All. In this mode, the "statistics" window
>>will keep all the best lines that the program has computed, as they arrive. The
>>window will scroll. So you will see exactly the time needed for every best line.
>>
>>This can be useful if you try a test position for example. You just set the
>>position, start the program's thinking, and go have a coffee. When you come
>>back, you know how much time the program needed to find the right answer, and
>>you even know if it has held the right answer all the time.
>
>
>I agree with your last point - I like to see if/when the best move changed.
>However, I still cannot tell how long it has been thinking for. So, for
>example, its last line may be for depth 14 after 1 minute, and it is now looking
>at depth 15 but not yet produced a pricipal variation for that depth. If I want
>to manually stop it after 5 minutes, I don't know when the time limit is
>reached. I know there are workarounds, but "time spend on current move" is
>handy to see.
Depending on the time mode you have chosen, you can see the time spent on the
current move in the status bar of the program.
>>>I like how it can automatically save the current position prior to exiting.
>>
>>
>>Me too. :)
>>
>>You can optionally tell the program to ask you if you want to save the position
>>or not on exit. I guess you are not going to change the default behaviour.
>>
>>
>>
>>>Optionally highlighting the last move is helpful.
>>
>>
>>Yes. I think we should have turned this option ON by default.
>
>
>I think this depends on your usage of the program. Torstein has mentioned that
>for a human playing against the computer, this can be annoying and it's a fair
>point. I like it because I play mainly comp vs comp and have a separate board
>setup. So, defaulting to "off" is ok - I'm happy that I can optionally switch
>it on.
>
>
>>>I prefer an evaluation score where negative always means "better for black". GT
>>>uses negative to mean that it is worse off, regardless of colour.
>>
>>
>>It's hard to satisfy every customer. Many people complain if we display the
>>score always from white's point of view...
>>
>>What should we do?
>
>
>I agree, it's not easy. I like Uri's idea of having this configurable so that
>people can change it to what they like best. This has the slight disadvantage
>that when people supply GT analysis they have to say what mode the evaluations
>are in. But it's still a good idea.
>
>
>>>The online help is poor. It is incomplete and vague. This was the most
>>>disappointing aspect for me.
>>
>>
>>Maybe Lex will confirm this: we are probably going to provide a free update for
>>the online help by Internet (not a big download).
>>
>>
>>
>>>Is it possible to reset the opening book learning weights? How should I use the
>>>dialog so that GT plays the opening as best it can? Learning? Variation?
>>
>>
>>Be sure to check Extra->Options->Books->Use engine books.
>>
>>When you do so, the program will play at its best. You will not be able to
>>change the other settings in the window, because the best settings will be
>>selected automatically.
>
>
>Thanks. This information is helpful.
>
>[snip]
>
>A few more minor comments/questions...
>
>GT doesn't ponder following a forced single move. I thought that it would
>benefit the program to use ponder as much as possible. I've also seen it not
>pondering when it is winning easily, but no mate in sight.
There are a few cases where GT does not ponder, but it's a minor issue. By
fixing these, how many elo points would the program gain? 1? 2?
I'll fix these problems, but I first focus on more important issues.
>How does GT handle 3 fold repetition during it's opening book? Does the "draw
>score" affect this?
No.
> Does GT learn not to play this opening again?
>
>Example using "ct" book, GT is Black against a "2000" rated player:
>
>1) d4 d5 2) Nf3 Nf6 3) c4 c6 4) cd cd 5) Nc3 Nc6 6) Bf4 a6 7) e3 Bg4
>
>8) Qb3 Bxf3 9) gf Na5 10) Qa4+ Nc6
>
>11) Qb3 Na5 12) Qa4+ Nc6 13) Qb3 DRAW
>
>
>
>Finally, please don't think I have major issues with GT. Most of my comments
>are indeed minor. As I said before, and most importantly, GT is a fantastic
>chess player and I'd recommend it to anyone.
Thanks.
Christophe
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.