Pouch Deposits The Editor and the Readership

In the last issue of The Diplomatic Pouch, Manus lavished copious praise on Andy for putting in so much effort for mastering the game DipPouch. After helping me out by formatting the Pouch Deposits and a column in HTML, Andy suggested I might want to do the praise thing again. But I thought about it for a while and decided that the amount of work he puts into the game is so monumental compared to a little HTML coding and that such lavish praise now would only take away from the praise received before. And I couldn't do that so such a nice guy, could I? (But thanks, Andy!). I added a few late arrivals, so any mistakes are probably mine.

In keeping with Manus' original idea for the Pouch Deposits, I'll bring up another topic for people to think about and reply to. Since as I write this my own article isn't finished, I won't be writing a column on the subject - I'll just throw it out for discussion and we'll publish any replies in the next issue:

The topic: play-by-Email Diplomacy vs. face-to-face Diplomacy. What do you see as the merits and disadvantages of one versus the other? Which are you more involved with? Which do you prefer and why? Feel free to discuss postal Diplomacy as well, though I supect that readers of this publication will be more heavily involved with PBEM and FTF Diplomacy than postal play. You can send replies using the "Dear DP..." form or by Email directly to me or Manus (see the bottom of this page for addresses and a link to the form).

I'd also like to take this opportunity to start a new addition to the Pouch Deposits section that will appear when there's material to put in it. It's called the Rants and Rave Reviews section and will contain kind words, positive feedback, criticism and lunatic ravings that we receive by E-mail. This issue's section contains one item from the first category and one from the last. Boy, I just love that second one. It's just close enough to English that you can tell it's a compliment for The Pouch, but not so coherent that you'd mistake the writer for somebody sane. That person perfectly captured the subtle line between English and jibberish, and walked it like it was a tightrope. Keep those rants coming. We can't promise to print all the ones we get, but I'd love to feature a few of the best ones.

- Simon

From Robert Rehbold (rehbold@dippy.franken.de)

You may remember that I complained about having only Lynx to read it and that I asked to provide the unlucky non-WWW-surfing world with a text-only ftp-able version. While I still think that this would be a good idea I have to admit that the DipPouch looses a lot by looking at its text only.

Looking at it with Netscape is like changing from hearing to viewing!

Congratulations on the great work which I hope to be able to enjoy a long time

[See the About The Diplomatic Pouch column for news on text versions of The Pouch. -Ed.]

From an Anonymous Reader (verbatim, honest!)

From Jonathan S. Tan (jontan@mit.edu)

(1) Tyrolia - used by Italy to attack Austria

As a beginner, I always thought that A Ven-Tyr and A Rom-Ven, followed by an attack on Austria, was Italy's best strategy. Even though this plan usually works in gaining Trieste and Vienna, Italy's mid-game is short-lived. With just Russia, Turkey, and Italy left in the east, Italy is naturally the next power to get eliminated for a number of reasons:

• Russia & Turkey have a natural division of centers which gives Russia Rum, Bud & Vie, with a continuing alliance.
• An IT alliance gives Turkey easy control of the east, where Italy will be the next after Russia to be eliminated. The IT alliance is good for Italy when it is necessary for IT to grab the eastern stalemate line to prevent a solo-win of one of the western powers (or to stop a western alliance from entering the east)
• The IR alliance is very difficult to maintain. Once Austria and Turkey are eliminated, it is almost natural for war to break out between Italy and Russia along their newly created long border of abundant supply centers.

In conclusion, my advice for beginners playing Italy is not to attack Austria 1901. The only time I have seen a successful early attack on Austria by Italy is when Austria was a gullible beginner and Italy suckered Austria to attack Russia and Turkey while marching into Vienna, Trieste & Greece in 1901! Another interesting consequence of the Italian opening of A Ven-Tyr and A Rom-Ven is that if Germany suspects that Italy will attack Austria in this way, and Germany wants to help Austria, then A Mun H followed by A Mun-Tyr will make Italy look rather foolish. Germany can then help Austria gain back anything that is lost.

(2) Tyrolia - used by the AI alliance to balance the western powers, provide defense, and allow flexibility.

Tyrolia can have a very useful purpose for the AI alliance in the beginning of the game. The immediate goal of an AI alliance is to eliminate Turkey while maintaining the status quo in the west. The opening move A Ven-Tyr (with A Rom-Apu and A Nap-Ion) achieves this in a rather subtle way. When the western powers ally against each other 2 on 2, then the power allying with Russia is at a disadvantage of one unit. Aha! The Italian army in Tyrolia can provide the balance and keep the status quo in the west while Italy and Austria attack Turkey. For example:

• If EG have FR outnumbered, then Italy's army in Tyrolia can be used to harass Munich or provide French support into Munich.
• If EF have GR outnumbered, then Italy's army in Tyrolia can provide support for Munich, or swing over to Piedmont to harass Marseilles.
• The only 2 on 2 combination that Tyrolia becomes useless against is FG vs. ER, but then Tyrolia can always be used to stab Austria or move up to Boh and maybe beyond to help AI attack Russia.
• If the western powers ally 3 against 1, well then you can just hope the alliance breaks apart and you can reestablish the balance of power.

The opening move of A Ven-Tyr is also very useful for defense in the AI alliance. It is good protection against an early western triple and prevents A Mun-Tyr and A Mar-Pie from gaining a supply center.

There is an even more subtle advantage of the opening move A Ven-Tyr. If Turkey does not trust Russia and orders A Con-Bul,F Ank-Con, and A Smy H, then Italy can forgo Tunis and Austria can loan Vienna with the fall moves F Ion-Aeg, A Tyr-Vie, F Con-Aeg, and A Smy-Con leaving Ank open for Turkey's build even when Russia and Turkey are allied! (The initial purpose of A Smy H is that if Russia moves to the Black Sea, then Turkey can build F Ank.)

Finally, if Italy and Austria are counting on Russia's help in attacking Turkey then Austria can order A Vie-Bud and the move A Ven-Tyr can help Austria defend Vie and Bud in case of a Russian stab with A War-Gal.

(3) Tyrolia - key to the last eastern stalemate line in the endgame.

Tyrolia is the key to the last eastern stalemate line (all 17 eastern supply centers minus Mos, War & Tun). Oftentimes, one or two eastern powers must jump to the stalemate line or grab the stalemate line when a western alliance is attacking the east. Tyrolia is one of the key provinces in this battle. If the western alliance can take and hold Tyrolia, then the eastern powers are without a stalemate line for their defense.

So, in conclusion considering (2) and (3), Tyrolia can be a very strategic province in all phases of the game.

From Simon Withers (swithers@under grad.math.uwaterloo.ca)

In one face to face game in which I played, this opening was used with the French knowledge and support: the French opened A Marseilles Burgundy, A Paris - Picardy, and F Brest - Mid Atlantic Ocean. The English had convinced the Germans to attack France. The Kaiser nearly went into fits when in the Fall of 1901, the army in Wales was convoyed into Denmark and the French had two armies breathing on the German interior.

Guest copyboy Andy Schwarz responds: How the heck did the Welsh army get into Denmark? Assuming the German opened with the anti-French F Kie-Hol, oughtn't A Wal-Eng-Nth-Den have bounced with A Ber-Den? Did the German make no move at all on Denmark?

Exchange Between Danny Loeb (loeb@delanet.com) and Simon Szykman (simon@diplom.org)

DL: It has come to my attention that the "refined evolutionary search" described in my S1995M DPP article in DP is a special case of the technique of simulated annealing used in artificial intelligence.

SS: Did you mean genetic algorithms? I'm pretty familiar with both genetic algorithms and simulated annealing, and what you described in your column sounded more like the former than the latter

DL: Tell me a little about both and I'll let you know.

SS: Okay. MAJOR simplifications here, just to give you the idea.

With Genetic algorithms, you encode a state in a simple representation (such as but not necessarily a binary string). You have a population of a number of such states. You then evaluate them according to some fitness function that says which are good and which are bad. You have two different operators which act on these states. A recombination operator takes pairs of strings, splits them up and recombines them into two new strings which replace the previous two (i.e. the two parents have passed on their genes to a new generation). The likelihood of a state being chosen as a parent depends on it's fitness function value, so that the higly ranked (most fit) have a higher probability of passing on their genes. As this process repeats, the average fitness of the population because the most highly ranked are the ones that tend to contribute to subsequent generations.

The second operator is a mutation operator, which takes some aspect of a state (such as one binary digit in the binary string) and tweaks it (with a binary string, a 0 would be changed to a 1 or vice versa). This introduces a random element in the process. Statistically, mutations that improve fitness tend to get passed on and those that reduce fitness tend not to, because the probability of passing on genes is based on the fitness of each state.

Simulated annealing: you have an initial state and a set of operators that can perturb that state to give you a new state. You also have an evaluation function that lets you evaluate them (the fitness function above is essentially an evaluation function as well)

The algorithm is as follows. Apply an operator at random and evaluate your new state. If it's better than the previous one, you automatically accept it. If it's worse, you may accept or reject it according to some probability. This probability is a function of a parameter called temperature. The temperature starts out high and decreases with time, and therefore the probability of accepting a step to a worse state starts out high and decreases with time.

Genetic algorithms are based on evolutionary concepts. The population improves because the best designs contribute to subsequent generations and inferior ones tend to die off. Simulated annealing is based on the annealing of metals, in which a metal is heated to a high temperature and is slowly cooled to eliminate internal stresses. The probabilistic behavior of the state in simulated annealing is similar to the motion of molecules in metals in real annealing.

DL: Very interesting. Funny thing is I guess I was aware of the work on genetic algorithms done by some graduate students while I was an undergrad at Caltech, but I seem to have not made the connection. (They were working with Turing machines.)

In any event, my "Refined Evolutionary Search" is essentially Simulated Annealing (except that I didn't call the mutation rate "temperature"). My Strategy Finder does not consider more than one set of orders at a time for any alliance.

On the other hand, your genetic algorithms sound like an even better idea. Especially when searching for moves for a large power/alliance. When doing recombination, I would tend to keep orders for neighboring units and coordinated orders "together". That way a good attack from one set of orders and a good attack for another, could be combined to make a set of orders with 2 good attacks.

From Marc Leotard (leotard@fusl.ac.be)

The other is about the game Fontenoy, in Issue 300 of Chapter Two (1992). Can you tell me whether this is still available, and where? I do love the Chaos variant. It reminds me of Middle Ages history, when feudal regions were slowly welded into national states. Is that what your article is about? If not, I'll write one!

Danny Loeb's response: I can't find issue 300 of EPC2. Please tell me if you can find it. (Manus, could you put out a request for people to find these back issues... Thanks.) My article was an End of Game statement for the first Chaos game run on a judge. Why don't you write such an article, I'd love it. I'd also like to see a survey of the various Chaos game played so far on the net.

Marc continues: I used to make computations about the number of possible openings in a standard game of Diplomacy. I soon abandoned the task as a hopeless waste of time. Most such 'openings' had no diplomatic pertinence, and I had no objective criterion to decide which were to be incorporated, and which not. I think the first thing to do is make clear what you are looking for. I did it in a game-theory setting, to build the payoff matrix of the normal-form game.

Danny's response: I suppose you saw the article (reprinted from EP) about the total number of openings for each country. I counted 2 openings as being different if there was the possibility of some moves by the other 6 players such that the 2 openings would lead to distinct final positions. This number is just a matter of trivia, so I didn't eliminate any "stupid openings".

The normal form of the game of diplomacy is even bigger than what you estimated. The payoff matrix is a 7 dimensional matrix indexed by different strategies. A "strategy" is a rule which says what a player does in every possible legal situation (the number of military situations was counted in another EP article, and is a REALLY huge number, the number of situations including the history leading up to it is of course infinite). The extensive form of the game is a type of tree structure. This is probably what you are referring to.

Marc continues: Problems were:

• How to measure the gain in the long-term.
• How to incorporate the negotiation element in the strategy vector.
• How to estimate appropriate subjective prior probabilities? Manus Hand's (is this a pun, or what?) proposal to compute the number of A01 positions seems easier; but the use for a game theorist is naught.
• And... are you French or American (or some place else)?

Danny's response: In the normal form of the game, the gain can be measured by the end result, and translated into a number using for example the hall of fame's scoring system. However, what you want to indicate is the approximate heuristic value of a position AFTER ONE MOVE. (Just like chess programs use static evaluation on the leaves' of their search tree, you propose estimating the position after one move for the purposes of creating a game matrix.)

My Bordeaux diplomat contains a static evaluation module as part of the strategy finder. However, you should be aware that even very small changes in the estimated values of the resulting positions may cause huge changes in the Nash equilibria (optimal strategies). (Barney Pell and I observed similar behavior while analyzing the game Stratego.)

I'm not sure what you mean by strategy vector. Of course, there may be multiple Nash equilibria. Negotiation will hopefully allow a player to choose a strategy corresponding to a favorable Nash equilibrium.

I'm American but I live in France

From Simon Szykman (simon@diplom.org)

From Markus Kaessbohrer (kassbohr@WHU-Koblenz.de )

1. Have everyone give orders for the MM, and select one set at random every turn (increases playing time, though). Nobody can be sure of an MM suicide then.
2. Have everyone give orders for the MM, and select the set that will result in the MM occupying the most supply centers after adjudication, or failing that have its units adjacent to the most supply centers. This requires experienced players, though, and might be helped by computer support

1. Yes, but the MM rules I described have the advantage that everyone knows whom to conduct diplomacy with on each turn and also the very interesting and deep subplots of trying to make sure that this or that player does or doesn't get control of the vacant power for a certain phase.
2. Ooh! I like this. Great idea! Actually, it would be as if the Game Master were playing the vacant power, and he only writes the orders for these powers after knowing what the other powers are doing. (Well, that's going a little far - or is it - will this be a guaranteed win for the Master? I actually think not. Certainly worth a playtest!)

Well, there you have it. Another Deposits column. Yep, it sure is.

Simon Szykman, Guest Editor, (simon@diplom.org)
Manus Hand, Publisher, (manus@manushand.com)

If you wish to e-mail feedback on this article to the author, and clicking on the mail address above does not work for you, feel free to use the "Dear DP..." mail interface, which is located here....