This came out of a discussion with Marius.

It seemed dumb to me at first but I dare ask it here all the same:

- should you play the best move considering only the dice and possible future moves, or

- should you play the best move according to your opponent's style?

- are there any bots who play the opponent instead of the dice which should be possible considering LG stats available?

Interesting topic. I think that you should always make the better move. And you should assume that your opponent will play his best, no matter which rating he has.

I think that bots behave like this, too. (I know that this is not an original answer!)

a) Yes, you can get a better performance if you use the weaknesses of your opponent. But it is not so easy to get knowlegde about his weaknesses.

b) If I understood it right, there is (was?) at least one programm not playing always the “optimal” move. This program gives bonus points in situations where it knows the game-theoretical value of a position exactly. So it prefers situations, where itself can play optimal.

c) During my long time of play with and against hanfried_c I spotted several times the following situation. Move A is better than move B. But playing A leads to an “obvious” position, i.e. both players know quite exactly which moves are optimal for the rest of the game. Playing B leads to a position which is simple enough that hanfried can compute quite exactly the perfect moves, but complicate enough that humans can make mistakes.

So playing B can be “better” for hanfried, if the difference between the winning probabilities of A and B is not to large.

d) Such situations are a part of the idea behind Ingo Althofers “3-Hirn-Prinzip”

(3-brain-principle). A 3-Hirn contains of a computer computing two “good” moves and a human, choosing one of them. Such a 3-Hirn plays better than its parts(*). In some games (chess to name one) amazingly better.

(*) The human need to know “something” about the game and the strengths and weaknesses of the program, but he can be – playing allone – quite weaker than computer.

Kitaktus

It is similar to Poker:

When you are good, you try to play perfectly (in a game theoretic sense).

When you are very good, you also try to model the opponent.

When I prepare for a Ewn match here on LG, I typically

play through some dozen games of the opponent to get

a feeling for his or her special style.

Ingo.

It's indeed similar to Poker:

The optimal strategy is the Nash-equilibrium, this means if anyone differs from that strategy his expectation value to win will drop.

BUT if somebody differs from the optimal strategy and you know how, then you can exploit this weakness and get an even higher win expectation than when you would stick to the 'optimal strategy'.

I would be interested in hearing from anyone who could comment critically on my style. And on a sampling of other styles, and what to do about them.

Hello Dvd Avins,

for my analysis you would be too strong, with your 1700+

rating. My analysis I only make for players below 1600.

Thanks for your interesting answers.

I was reasoning along this line of Nash equilibrium (see Footstep thread: http://www.littlegolem.net/jsp/forum/topic2.jsp?forum=1&topic=1730), but isn't EinStein a game of perfect information contrary to Footstep ? I'm not sure the concept of Nash equilibrium thus apply here.

However I think some mistakes may be exploitable, the most common being players who avoid to take their own pawns enough. But how would you exploit this?

> However I think some mistakes may be exploitable,

> the most common being players who avoid to take their own pawns enough.

A second common weakness is that players play too defensive

(against Theo van der Storm's advice: “always be in a hurry”)

> But how would you exploit this?

One has to look at concrete examples…

When I stumble across one, I will post it here.

Ingo.

Against this weakness (too defensive), I guess the opponent prefers to take your advanced pawns rather than his own, I think you have to advance more pawns instead of taking them in order to let the opponent eat a lot until you have very fast pawns?

Here is an example with a player that has a tendency to take too many of the opponents paws too early, I really don't know how to exploit in this position:

[game;id:1357906;move:23]

As I don't have so many pawns left, it will not be an advantage to jump in the battlefield…

who plays?

I just saw that this game is not over, yet.

Please, no examples from running games.

Ingo.

Yes you are right we can discuss after I've lost…

It was just to remember this interesting position, Red to play and win in 6!

I haven't seen this thread before so I have to ask:

@Carrol: Marius who? Me Marius or my stronger xará Halsor? I mainly ask because I commonly discuss EWN positions but I don't recall doing so with you.

Yes MRFvR, Halsor is his name, I did not know you were Marius too, but please discuss the game next time we play.

Here is his analysis he shared in our game and which I find a gold mine:

“In EinStein, I sometimes try to go around the edges, sometimes I fight for the center, sometimes I try to annihilate my opponent and other times I try to eat my own stones to gain more speed for important stones. All depening on the current situation. Of course, I still don't play perfect, so an “exploitive” strategy will still be better than another strategy with equal win-probability given perfect play from your opponent. But I wouldn't be willing to reduce that win probability in order to gain an edge if my opponent makes a mistake, I think. At least I'd only be willing to pay a very small price, and only if the possible gain was considerable. "

Marius tell me if you object to share this wisdom, I'll remove this post…

Not at all, Carroll, feel free to post my comments here. I'll make sure to let the recipient know if something I say is said in confidence :-)

Thanks!

Btw I lost this set to kiwibill around this move:

[game;id:1357906;move:35;title:About 77%]

One example of opponent modelling: When the opponent

is a bot which uses complete 6-stone endgame bases,

it may make sense to avoid too many captures in

the early game.

Ingo.

On the other hand, when you are a bot with 6-stone

endgame bases, you may capture some more pieces than

usual against opponents without databases.

So, trying to bring the game in an environment where

you are perfect.

That is interesting, how big is the 6-stones database and does it really give an edge on bots which can look at the tree with much depth when only 6 pieces are left?

I'm not sure I agree that a bot should lose a few percents to get to positions where human beings would play better than positions with many pieces… I think that in most games, bots should go for a more complex world, for example in chess play a close game.

Years ago, Ingo Schwab had made some analysis on EWN,

and later also Jakob Erdmann in his doctoral dissertation.

Most difficult decisions in EWN (and greatest chances for

human blunder) are in the late middle game.

Ingo.

@Carroll: If only 6 pieces are left, the database is not really neccessary.

But it is verry helpful if you are in the middle game and you can easily look up the game theoratical values of some thousands 6-stone-nodes in your search tree.

@Ingo: I would say the greatest chances for human blunder (looking at a single move!) are in the early endgame. But perhaps this is a question of definition.

If you give me a list of “interesting” games, I could do an analyse.

Thanks for the kind offer.

I would like to see Hanfried's comments on this game:

http://www.littlegolem.net/jsp/game/game.jsp?gid=1362662

Especially, there would be a chance to get (later) also

comments from Sybil_c.

Ingo.

@Ingo:

The first 3 games are analyzed. Sybil is playing nearly error free (as far as I can see), most of the time. But move 41 looks absolutely strange.

I will send you the details in an email.

Thanks, Hanfried.

> @Ingo:

> The first 3 games are analyzed. Sybil is playing nearly error free

> (as far as I can see),

And what about the quality of my moves?

> most of the time.

That reminds me of Richard Lorentz. We had done a quick trip on

German Autobahn (A4, from Jena to Gotha), me at the steering wheel.

Our maximum speed was 160 km/h, so not overly fast.

When I asked at the end “Did you feel save?“, Richard replied

“yes, most of the time”.

> But move 41 looks absolutely strange.

Interesting. That might be a consequence from the special structure

of the bot (as described in Sybil's info text). Other participants

in the Olympiad might try to exploit this weakness.

> I will send you the details in an email.

Thanks in advance, Ingo.

PS. Later Richard specified: He did not feel completely well

when we were on the middle lane with about 150 km/h,

and were overtaken by other cars who were much faster.

overtaken from the right?

> overtaken from the right?

No, properly from the left. Everything was correct.

And especially, we were not sandwitched.

Before entering Autobahn I had made a sketch for Richard…

“Here you have the three lanes: on the rightmost cars typically drive

between 90 and 130. On the middle lange normal speeds are

between 120 and 150. And on the left lane most cars are between

150 and 250 (the majority of them below 200).

When entering Autobahn you have to make sure that acceleration is

fast enough. So, I will go in second gear until we have 80 km/h,

then in third gear, until 110 is reached.” And in the presentation

it worked really well, although the entry at Apolda (on A4, in direction

to Gotha) is not completely trivial.

Ingo.

Richard Lorentz here. (I believe I'm one rated game away from becoming trustworthy and thus being able to post under my real name. It's your move, Diamante. :) )

Here in the U.S. (at least where I drive, around Los Angeles) the fastest any reasonable person drives is around 140. Of course one needs to get out of the immediate L.A. area to have any hope of going even that fast, but that's a whole other subject. Anyway, cruising at 160 was a bit uncomfortable being that it was as fast as I had ever gone in a car. One gets the sense of instability at that speed. Then, when another car passes going quite a bit faster, I could feel the resultant air turbulence giving me the sense we were on the edge of disaster.

Fortunately Ingo is a very good driver so I wasn't that worried – most of the time. :)

-Richard

A few more bits of information on that Autobahn tour.

In the web I found a picture of

Autobahn A4 - Entry Apolda.

The blue truck is in the direction where we drove. The entry is a bit

tricky because it is in a sort of valley. The trucks (100 km/h allowed

for them) come downhill typically with 100 + (5 or 10), to have enough

momentum for the uphill. Especially, the drivers are not always so

prepared to brake for “entrepreneurs”. You see, entering itself is

nontrivial (there are lots of trucks on the A4). And, when you want to

do it well you yourself have to be fast enough to reach the middle lane

soon. Otherwise you will get stuck behind a truck, when they slow

down to 80 or even 60 in the uphill phase.

On the evening with Richard things went well. Probably, he even did

not realize that our task was nontrivial.

Ingo.