Human killer engine - a cafè-monster

[pocopito]

Just to get an idea of how many games can play an engine, this is the info for Kitteneitor, which I try it to be playing 24/7:
http://www.ficsgames.org/cgi-bin/search ... ;year=2013

For Kitteneitor I use a netbook that got its screen broken, so I got it for free, and it has a low power consumption.

Engines can play tournaments but I disabled this option on cutechessclient because it only was useful to play against other engines.

About resigning criteria, I have no idea on how to implement it on freechess.

Best regards

Emilio

[JBNielsen]

Just to get an idea of how many games can play an engine, this is the info for Kitteneitor, which I try it to be playing 24/7:
http://www.ficsgames.org/cgi-bin/search ... ;year=2013

For Kitteneitor I use a netbook that got its screen broken, so I got it for free, and it has a low power consumption.

Engines can play tournaments but I disabled this option on cutechessclient because it only was useful to play against other engines.

About resigning criteria, I have no idea on how to implement it on freechess.

Best regards

Emilio

Thanks for the insight and your advices.
Good inspiration.

DabbaFun is ready now for the next 12 hours.

[pocopito]

Anytime.

By the way, I just remembered that I made a couple of tries with a raspberry-pi, but it disconnected continuously. I did not try it really hard, so maybe I was making some kind of mistake, but I guess this kind of platform is almost perfect for your needs, specially taking into account that you are not interested in performance but in developing a "fun to play against" engine.

Another maybe useful hint. In order to lower energy consumption maybe it could be useful the cpulimit application (if you work under linux), that allows you to limit the amount of cpu a certain process can consume. I haven't used it for a long time but a couple of years ago it seemed to work pretty fine.

Best

E Diaz

[Rebel]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

It is a bit like that I want.

What helps is to add a bonus to root-moves that leave pieces hanging ranked by piece-type multiplied by the number of hanging pieces. The effect is obvious, it complicates positions for the human brain.

[JBNielsen]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

It is a bit like that I want.

What helps is to add a bonus to root-moves that leave pieces hanging ranked by piece-type multiplied by the number of hanging pieces. The effect is obvious, it complicates positions for the human brain.

Good idea.

Could it be better to do it deeper in the search; eventually in the evaluation function?
It takes more time, but it probably works fine.

What about including pins?

[Rebel]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

It is a bit like that I want.

What helps is to add a bonus to root-moves that leave pieces hanging ranked by piece-type multiplied by the number of hanging pieces. The effect is obvious, it complicates positions for the human brain.

Good idea.

Could it be better to do it deeper in the search; eventually in the evaluation function?. It takes more time, but it probably works fine.

I have it in eval as an option, a match showed a loss of approx. 10 elo. Which ain't bad considering the goal. A variation is to give the bonus during search only for the odd plies, I am testing this right now, you got me started Another idea I want to try is to penalize (root) moves for winning captures (SEE>0). The effect (I hope) will be that such winning captures will be delayed until the search is forced to capture, thus complicating the position on the board.

What about including pins?

I have a parameter to increase the bonus for pins, I will try some crazy values and see what happens.

[JBNielsen]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

It is a bit like that I want.

What helps is to add a bonus to root-moves that leave pieces hanging ranked by piece-type multiplied by the number of hanging pieces. The effect is obvious, it complicates positions for the human brain.

Good idea.

Could it be better to do it deeper in the search; eventually in the evaluation function?. It takes more time, but it probably works fine.

I have it in eval as an option, a match showed a loss of approx. 10 elo. Which ain't bad considering the goal. A variation is to give the bonus during search only for the odd plies, I am testing this right now, you got me started Another idea I want to try is to penalize (root) moves for winning captures (SEE>0). The effect (I hope) will be that such winning captures will be delayed until the search is forced to capture, thus complicating the position on the board.

What about including pins?

I have a parameter to increase the bonus for pins, I will try some crazy values and see what happens.

Great with your work!

Can you tell us more:

* Human options should be tested against humans. Do you have your engine running somewhere?

* Another good idea to delay winning captures! Could it be better to delay ALL captures?

* you wrote "a loss of approx. 10 elo". But is it interesting to see how these human options affect play against other engines?

* have you plans to go on with the REAL human stuff; the swindle function? If so I can offer help with test positions, test, ideas... Instead of trying to make it myself in DabbaFun (which only scores 17% against the normal Dabbaba, by the way)

[JBNielsen]

Just in case anyone is interested on following its results and games:

http://www.ficsgames.org/cgi-bin/search ... Statistics

Regards

E Diaz

DabbaFun now has a rating of 2111.
But it is only based on ca. 40 games.

I have let it challenge many players.
But only very few of them come back and play new games by themselves

[pocopito]

Maybe human players find it too tough to play against it.

Another possibility is that the time control is a bit too long. Maybe you can try a time control ilke 1+2 or 1+4, and check if this way DabbaFun gets more opponents.

With this time control and an elo around 1900 Kitteneitor does not have many problems to be challenged.

[JBNielsen]

Maybe human players find it too tough to play against it.

Another possibility is that the time control is a bit too long. Maybe you can try a time control ilke 1+2 or 1+4, and check if this way DabbaFun gets more opponents.

With this time control and an elo around 1900 Kitteneitor does not have many problems to be challenged.

Thanks for the advice.
I have now set the time+inc to 1+4.

A faster game should be in favour of the computer