Looking for a Random Mover

[Kirill Kryukov]

Perhaps Olivier might help here. Have a look at http://www.open-aurec.com/chesswar/Chesswar015/Chesswar015PSt.htm
These are the lowest entries of the promo-division (all below Youk):

(list skipped)

I am not sure which engines are freely available though. (Olivier could help you with that one I guess).

Stephan

Hi all

All of them but Etabeta are freely available. Some of those ones are very buggy though and I suppose Kirill is looking for stable engines.

From this list the stable ones are (if my memory does not fail) Youk, DeepDuke, Geko, NEG, Pyotr, Mainsworthy (and the other engines by the same author), ACE, Akiba and POS. They may lose on time if the time control is a bit fast, but at least they won't crash.

Olivier

Hi Stephan, Oliver! This is very nice, and yes, I will gladly use this list and any other help I can get to collect useful weak engines. My project is very slow, so I'm not in a hurry. I will ask now and then if I am missing a particular engine or need advice. Thank you for being so kind to me!

Yes, stable engines are the ones I am particularly interested in.

[Kirill Kryukov]

Many of those engines are not usable. At least not without altering the underlying theory about rating, and the extraction procedure that follows from it. Some of these engines are so weak because they forfeit the majority of their games on time, and would do that irrespective of how weak the opponent is. When they are lucky enough not to crash, they can play quite good. So they might score 25% against a 1200-Elo opponent, 25% wins and 75% forfeits. But when you would pit them against a random mover, (say that has 0 Elo on the same scale), they would still score the same 25%. Scores between engines like that will not behave according to the Elo curve, so that it is questionable if Elo rating even is a meaningful concept for them.

Right, there are many problems with weaker engines. I manually check all crashes and time losses, and only use them if the crashed/lost side was already in lost position. It may be not a perfect solution, but the best I could come up with.

Still some engines may behave strangely, and I try to detect those by checking their "performance slope". An engine that loses 75% of its games regardless of the opponent will have a steep performance slope, easy to detect. For an example of similar engine, look at Chispa 4.0.3 at KCEC site. This engine is unable to win large percentage of won positions. Instead it makes draws. So it has enorumous draw rate, and steep performance slope (Perf = C + 0.266 [±0.105] * Opp). Similarly I will be able to detect other buggy engines, and in extreme cases I can remove them from the list.

[H.G.Muller]

What I worry about is that you would have to remove nearly all of them from the list. Absence of repetition detection and determinism are more rule than exception amongst these engines. Even micro-Max 1.6 suffers from this, which makes it very difficult for it to win games even against weaker opponents.

The problem is that almost all these engines are attempts at a serious engine, in which the design would guarantee a rating of about 1800 if they were not seriously buggy somehow. Only a few play genuinely weak Chess, e.g. because the author limited them to a low number of plies. The rest is extremely erratic in its behavior.

[Stephan Vermeire]

Perhaps we could 'generate' some weak derivates from the stable engines we have. If it is for obtaining a good ELO-scale, it might be usable. I could easily generate a Brutus-version which has no more evaluation settings than material. I also could switch off quiescense and set the maxply to 1 or 2. What do you think? Would that be a good approach?

Stephan

[H.G.Muller]

Putting severe limits on the number of ply, or better yet, nodes an engine was allowed to search would indeed be a good way to create a weak engine with non-sick behavior. It would be difficult to weaken them to the level of NEG, though. With 1 ply + QS you would already be vastly superior to what NEG does. But at least I can imagine you could get into the realm where a deep-searching random evaluator would probably be.

[Stephan Vermeire]

Well, I know what 'NEG' stands for! ("Niet Erg Goed" or translated: "not very good"). That is why I suggested to switch off Quiescense. My first engine that I ever wrote didn't have a QS-function and I can tell you that is was bad!

If you are interested: http://www.xs4all.nl/~vermeire/chess2001bc.html. It is a Java-engine that can be operated from the browser only (no winboard, timecontrol, garantee that your PC will survive...)

Stephan

[H.G.Muller]

This is fun!

I changed all parameters of the micro-Max evaluation to zero, and removed the QS (i.e. it only stands pat in QS, but never tries any moves). I also removed LMR and check extension. I left in null-move pruning, because it doubles as in-check detection, and is thus needed to distinguish checkmate from stalemate. For the eval I use the lowest 15 bits of the a hash-key-derived value. I called this engine "Ram 2.0" (a contraction of RAndom Mover).

This really plays like an idiot, but increasing the search depth seems to help. I equiped it with an option "Depth Limit", that can be set between 1 and 20. (It will never reach 20, I guess, except perhaps in KPK.) This is different from the WB st command, the effect of which is reset before each new game; when you set this option to, say, 3, Ram will be limited to 3 ply as long as it is not restarted, or until you change it again. I added this so that you can easily set a different depth for both engines, if you play Ram against itself: just use the respective Engine-Settings dialogs to give each engine its own Depth Limit.

For those who want to have a good laugh, the engine can be downloaded from http://home.hccnet.nl/h.g.muller/ram.exe .

[Roger Brown]

This is fun!

SNIP

This really plays like an idiot, but increasing the search depth seems to help.

SNIP

For those who want to have a good laugh, the engine can be downloaded from http://home.hccnet.nl/h.g.muller/ram.exe .

Hello H. G.,

I will try out your engine but I will be the judge of what plays like an idiot. I am extremely well suited in that department.



Too often we have been cruelly decieved by engine authors who proclaim the weakness of their engine only to find out you were playing a 2000+ Elo monster.

Later.

[Edmund]

Any news on this interesting topic? Are the random mover tournaments already under way?
regards,
Edmund

[Kirill Kryukov]

I recently saw BrutusRND participating in Adam Hair's rating list: Link. It's the lowest rated engnie, as expected. 103 points lower than POS 1.20.

Unfortunately I haven't been able to include BrutusRND into my tournament yet.

[Edmund]

I recently saw BrutusRND participating in Adam Hair's rating list: Link. It's the lowest rated engnie, as expected. 103 points lower than POS 1.20.

Unfortunately I haven't been able to include BrutusRND into my tournament yet.

Thats nice. So this would suggest the Elo Ratings are about 450 Elo Points off your more natural scale.

Anyway, there is another issue I wanted to address. Plying strength has also to do with the timecontrol used. So if you want to use the Random move as a Base 0 Elo engine, and you add another engine which is around 500 elo stronger, what TC is this based on then? The Random mover doesn't gain much Elo by thinking longer, the stronger engine might well improve. So it feels to me, whenever someone wants to quote an elo with your base he has to specify the Timecontrol (and for computers also the Processing Power). And as you move up the rating scale, playing many games against the 0elo engine with a certain stronger engine to get a new base engine, you would actually need an elo rating for each common timecontrol.

Any thoughts on this?

[Betsemes]

Hi.
I'm also looking for a random mover, and I found this link: http://home.xmsnet.nl/vermeire/BrutusRND.rar somewhere. Unfortunately, it seems to this website is blocked in my network. Could anyone email this file to me? If so, please, PM me so that I can send my email address. Thank you.

[chessp]

i wanted to use brutusRND to experiment with endgames against untrained players (which is sometimes surprisingly hard). but it thinks too long. why does it have to think half a minute for a random move?

[H.G.Muller]

Well, use POS then...

[chessp]

this thing is fast, but unfortunately it's the oppsite of random, it does the same thing every time! is there no fast simple random mover, even today? i understood prior posts to mean that brutusrnd does it, am i doing anything wrong?

[H.G.Muller]

You did use POS with the option --brain-mode random?

[chessp]

another problem with pos: i cannot save a game where pos was black. when clicking save game i get error code 12290, which seems to indicate that some path or filename is illegal. winboard seems to put the names of the players into the suggested filename, but it does not get them from the tags. and the pos opponent calles itself (or is called by winboard) "POS v1.20 / Mon Dec 29 12:10:19 2008", and this slash kills the filename.

where do the automatic player names come from and how can i change them?

this is also a bugreport/feature request for winboard: run some string cleaner over any string which could become a filename. or in case of error fall back to a filename like "new".

[chessp]

i did not. now i pasted it into the "edit engine list" window, which resulted in a fatal error complaining about "unrecognized argument --brain-mode". what to do next?

(un)related question: i deleted brutusRND, how get i remove it from the bottom of the engine menu?

[H.G.Muller]

The name is supplied by the engine itself. There already is some cleaning up of the name when it is used to propose a filename for saving (e.g. spaces in it are replaced by underscores), I guess I just have to extend that to (back)slashes as well. I suppose the Windows file-browse dialog does not even pop up, so that you don't get a chance to change the the name into something that is acceptable to Windows?

I guess the work-around for now is to specify automatic saving of games in the Save Game Options dialog; there you can select whatever filename you want.

As to the --brain-mode thing: are you sure you added this as an option to the engine, and not as an option for WinBoard? The engine line should look something like

"pos.exe --io-mode xboard --brain-mode random" -fd POSFOLDER

I must admit that I have never used POS 1.20; my experiences were with an older version (1.17?). So it is possible that '--brain-mode random' is no longer supported in 1.20.

As to the engine menu: you mean the recently used engines displayed ther? Good question. These are stored as an option -recentEngineList in the settings file (which lives in the user's AppData folder, which is usually hidden). I suppose you deleted it from the engine list already. The menu contains the nicknames of the engines in that list, which is then used to locate the full engine command in the list when you click the item. I guess it would be nice when, in case the nickname could not be located, the corresponding nickname would be removed from -recentEngineList. Of course sooner or later the name would disappear by itself, when you don't use the engine anymore, but do use others.

The easiest way to get rid of it now is to run WinBoard with 'Additional options': -recentEngineList ""
This would clear the entire list.

BTW, if it really doesn't work with POS, I could convert my non-searching engine N.E.G. which I puplished on TalkChess last month easily to also support a purely random mode. It doesn'y know how to castle or e.p. capture, though; I don't know if that is a problem for your application.

[chessp]

yes, the dialog doesnt even show up. inserted an autosave filename now. when and where does it do the autosaves?

in engine/edit engine list i add the new parameter to get the line:
"pos-1.20.exe" --brain-mode random -fd "C:\Users\Peter\Documents\_synceddocs\chess\downloads\pos-1.20"

yes, recent engines. to have it disappear automatically is not an option since i only use fruit and stockfish. i will go hunt for that settings file.


and it gets more off topic: is it possible to launch winboard (maybe with a .bat file) to automatically load a game from a .pgn file and load an engine and assign this engine to black and then let me start playing as white?
right now i launch the pgn file, double click the game, load the engine, make the first move as white and then press ctrl-B to get stuff going. not very smooth.