Collector's Corner : Bizarre loss on time

[JonP01]

I've been playing computer versus computer games for many years now, but I honestly cannot remember the last time I had a computer actually lose on time. Infact, I am not sure it has ever happened before today. I have always followed the policy of allowing a machine to calculate the time left via it's own clock - this way my operator time does not factor into it. For example, the Mephisto Miami will count down it's clock while it is thinking and stop it's clock when it's move is made. And whatever opponent it is playing - the same rules applies.

Today I played a game between my Excalibur Deluxe Talking Chess and my Chessmaster Gameboy. Although Chessmaster is proving to be the stronger machine, the games are often quite interesting tussles - there are usually fireworks right off the bat - in which case Deluxe talking usually suffers, or they are longer sluggers or the game simplifies early - in which case Deluxe Talking has managed a couple of draws.

In today's game, Deluxe Talking was actually in a slightly better position, but for some bizarre reason her clock ran out of time on the 33rd move. Prior to that, on white's 31st move, she thought for a whole 8 ply. That is 2 ply more than normal, although in complicated middle games she may do 5 plys and in sparse middle games might just start the 7th. But 8 takes much longer. by the time she had completed her 31st move, she only had a few minutes left to complete the remaining 9 moves. Thinking she would speed up to make up for time, she took well over 2 minutes for her next move. And on move 33, her clock counted down...to zero.

What makes this even more bizarre is that the move where she lost all that time calculting 8 plies - there were only 2 legal replies to consider!
As I say, I was quite amazed this happened, although I have noted that one side often loses on time with this machine if it plays itself.


[Event "?"]
[Site "?"]
[Date "2007.04.05"]
[Round "?"]
[White "Deluxe Talking Chess"]
[Black "Chessmaster GBA"]
[Result "0-1"]
[ECO "B08"]
[PlyCount "66"]
[TimeControl "40 in 2"]

1. e4 g6 2. d4 d6 3. Nc3 Bg7 4. Nf3 Nf6 5.
Bc4 Nc6 6. d5 Na5 7. Be2 c6 8. O-O cxd5 9. exd5 O-O 10. b4 Nxd5 11. Nxd5 Bxa1
12. Bg5 Bf6 13. Bxf6 exf6 14. bxa5 Be6 15. Qd4 Bxd5 16. Qxd5 Re8 17. Nd4 Rc8
18. Rb1 Rc7 19. a6 bxa6 20. Bxa6 Re5 21. Qf3 d5 22. Bd3 Kg7 23. Qg4 Qe8 24.
Nf5+ Kh8 25. Nd6 Qd8 26. Qb4 Rd7 27. Nb7 Qb8 28. Nc5 Qxb4 29. Rxb4 Rc7 30. f4
Re1+


[D] 7k/p1r2p1p/5pp1/2Np4/1R3P2/3B4/P1P3PP/4r1K1 w - - 0 31


31. Kf2 {White took an extraordinarily long time on this move. It
calculated to the 8th ply rather than the usual 6th ply. White only had a few
minutes left on it's own internal clock after completing this move} Rh1 32.
Rb8+ Kg7 33. Na6 Re7 {White ran out of time according to it's own internal
clock - white loses on time (black had used around 90 minutes)} 0-1

[Steve B]

Actually Jon
i Do notice this sometimes does happen even with other computers
although i never saw it with the DTTC

it could be a quirk in the code for search extensions or just that the computers freeze in some way

generally on the rare occasions when this occurs i just force the computers to move and i usually find they play to the time control fairly accurately from then on

of course i dont play the computers against each other to determine ratings or anything serious like that so i dont mind intervening in this way

3 Hirn Regards
Steve

PS.3 Hirn is a method of competing where two computers play a game against another computer with a human operator choosing the best move (in his opinion) from the two
if both computers choose the same move then this move must be played
the literal translation from the German is "3 Brains"
its actually quite fun and you should give it a try sometimes

[JonP01]

Hi Steve,

I am sure the problem is a bug in the program - perhaps this bug may relate to a parameter that adjusts the program according to the speed of the CPU or similar.

In any case, I note that DTTC loses on time with alarming regularity. Infact I now realise I have simply been "lucky" that the games I have played with it have never gotten as far as the first time control. Or in the cases where they have, they were rare instances of the bug not manifesting itself.

As I say, it is an easy bug to see. If the unit is simply left in "0 player" mode (so that it plays itself), one side will almost always lose on time under the tournament time controls. I thought this might have been a bug related to playing itself only, but it now appears that it happens exactly the same way if it is in "1 player" mode. Had I been formally testing this unit as a prototype, I would have found this problem a lot earlier than after a handful of games. It is rather unfortunate that the "budget" version of this program (LCD Talking) has the opposite problem - it moves far too quickly. I remember saying that whilst the LCD Talking did move far too quickly, it was better than losing on time!

I have to say that my experience with this DTTC unit has been incredibly frustrating. Not only is there the loss on time bug, but there are issues with matrix display corruption over time as well (and this has nothing to do with the battery type - I have subsequently confirmed that on an independant unit). Furthermore, the continuations it shows for the opposing side in "info" mode are garbage - it will often show illegal moves, horrible blunders or suicidal moves as the anticipated continuation for the other side. And then of course there is no permanent brain.

In the end, to an end-consumer like me it doesn't matter why the unit has these bugs - only that it does. I know that there may be issues within companies that cause products to be released without the thorough testing they deserve, however again - that makes no difference to me as a consumer of these products.

I'll go as far as to say that Excalibur products are becoming the bane of my computer chess hobby. I have a very long list of problems and significant bugs with every single Excalibur device that has passed through my hands. I think this is a good indicater that the dedicated hobby is never going to be anything remotely like it was in the 1980s and the early 90s. Back then, because dedicated machines still ruled the market, they were more thoroughly tested (against humans and other machines) than they are now. If the modern day Excalibur units were tested with the same thoroughness that typified the 1980s, then they would likely be a far better product than the ones being shipped to consumers today.

To make matters worse, Excalibur gives the impression that it is not necessary to communicate with it's customers and that emails sent to them can simply be banished to the "ignore" pile or the virtual rubbish bin. Would sid Samole put up with the quality of Excalibur's non-existent customer service and the endless stream of bugs with which their machines are released with? I strongly doubt it. I am starting to regret having ordered a GM machine now.

Hopefully you and others will understand that I am not taking a dig at Ron Nelson. I was a programmer in my professional life (financial applications) and I am fully aware that it is next to impossible to write a program that is bug free and works properly at the beta stage. I've never done it and I do not know of any programmer who has managed it (at least with a program of any degree of complexity). Even when I was sure a program worked perfectly, the phone would ring within a day the beta testing started with a user reporting a bug. And even if I had spent a week testing it myself, I might not have discovered that bug. It is critically important in testing to have a range of users with different focuses - only then is the chance of bugs making it through to the final release minimised. And it is critical to involve expert third parties in a testing process, rather than rely upon yourself as the technical guru to uncover all the issues.

So having said the above, I can't be sure why the products are coming out with so many bugs, but I feel that pressure to get products out the door according to a schedule might be a contributing factor. In other words, I feel it likely that if Ron Nelson had adequate testing support, testing feedback and enough time, no Excalibur product would ship with bugs of any significance, regardless of how pedantic the end-user of those products might be.

Still, I can't see things changing. That is the world we live in these days. the motto seems to be: Get it out the door and keep the shareholders happy. No wonder the old dedicated machines are considered classics and sell so well on eBay compared to stuff from the late 90s and beyond.


Relinquishing my Excalibur Fan Club Membership Regards


Jonathan

[Steve B]

Hi Jon

you will be very pleased with the GM

I bought two of them when first released and have had no mechanical problems with them to date

i never use them with batteries though..only the adaptor
in fact ..if a computer will run on an adaptor in addition to batteries i will always use an adaptor
unless of course i am taking the computer with me for some
"in transit" play

Stood Up In Cafe Regards
Steve

[klute]

Greetings

Having tested the GM more extensively in serious tournament play than any other on this planet, I hereby certify the GM as fit for purpose. Minor bugs sometimes manifest themselves, but nothing too serious - unlike many other Excalibur machines.

klute

[Steve B]

Greetings

Having tested the GM more extensively in serious tournament play than any other on this planet, I hereby certify the GM as fit for purpose. Minor bugs sometimes manifest themselves, but nothing too serious - unlike many other Excalibur machines.

klute

THE CAMERON PARLE!

well well
its about time you made a post

Humbled Regards
Steve

[klute]

Hi Steve

Yes it really is me - with eight machines blinking away on 40/2 at 1.29 am on Easter Saturday. Even as I write, GM contemplates its 51st move in game 11 of its match with Miami - will its extra pawn suffice for victory? Can GM go on to avenge a crushing 8.5 - 1.5 loss to Miami in the first ten rounds? Time will tell...

Easter Bunny Regards

Cameron

[Steve B]

Glad to see you have lost none of the fire in your belly for our beloved hobby

Play On my Friend..Play on ..Regards
Steve

[JonP01]

The Klute has arrived The most comprehensive, thorough, analytical, resourceful and knowledgeable dedicated computer tester on the planet (perhaps the Galaxy). I am sure I have met this guy somewhere before

Anyway, I do look forward to receiving a GM that will hopefully be fully functioning. It's never really been hardware issues with Excalibur machines - mainly just irritating program issues that frustrate because given more time pre-production, they could have been eliminated. Like the Klute, I like things to work properly, dislike having to make workarounds and don't wish to intervene as an operator because that represents a direct influence over what the program should be capable of deciding for itself.

With regard to DTTC, I have decided to compromise a little. Luckily, the device has a time control of 45 moves in 115 minutes. After playing a couple of test games against the Palm (to speed things up, since the Palm only needed 5 seconds per move and DTTC does not think on opponent's time), I noticed that DTTC ended up using similar amounts of time to the majority of machines when set to 40/2. For example, at move 40, it had used 1 hour 51 minutes.

In a second game, whilst it's own internal flag did indeed fall once again at the 45th move (time control), at the actual 40th move itself, it had used around 1 hour 55 minutes. These seem to be a worst case scenario, since it is more apt to run into time problems when in a losing position. And for some strange reason, in each instance I have tested, it spends an extraordinary amount of time when in check and there are two legal replies. Perhaps this was a coincidence.

So I feel using this time control, the chances of a loss on time without operator intervention are very much reduced, and the unit will use similar amounts of time to those it plays against. I have even made the decision that if it's flag is about to fall on the 40th move, I will hit the "move" button.

Work Around Regards

Jonathan

[JonP01]

Can GM go on to avenge a crushing 8.5 - 1.5 loss to Miami in the first ten rounds? Time will tell...

I was surprised to note that in the games you sent me, GM had a 15% score against Miami and a 25% score against Cosmos. I know it is a small sample of 20 games, but over at Scachcomputer.info, there is a similar result emerging (albeit a tiny sample):

http://www.schachcomputer.info/html/akt ... liste.html

Notwithstanding only 33 games, there does indeed seem to be an emerging trend that GM might achieve better results against Comsos / GK2100 than Miami / GK2000. I can't think of a reason as to why though. Perhaps more games might clarify the situation.