What is the best country in bot games?

[DinOdinson]

The bots don't know how to play as Austria, but if a player draws it, it's a winnable game if things break right. But you're also at solid odds to get hostile R/I/T opens and just die quickly, and there's little you can do about it.

[DougJoe]

An update, just because I'm in the mood. Apologies for the book.

I'm up to 19 games as each power against the bots, for a total of 133 games (results and stats, as mentioned before, at https://docs.google.com/spreadsheets/d/ ... sp=sharing)

For purposes of this discussion, I'm going to equate the question of "best country" to "who is easiest to solo with".

For me, count of Solos by power:T:12,I:11,F:9,G:7,E:5,R:5,A:3. If I look at Solos + Draws, it's T:14, F:14, G:13, I:12, E:9, R:9, A:7. My solo rate right now is 52/133 (39.1%).

Turkey: Extremely straight forward, with a solo rate of about 63%, almost 2/3. 17 of my 19 games have started with Con->Bul, Ank->Con, Smy->Ank. All 12 wins were in those 17 games, which is a win rate of over 70%. You may need a little help from Russia around Trieste, but unless there's a very strong A/I going on (and given bot Italy's proclivity to attack Austria, it's rare) you've got a good chance and timing the stab on Russia is the main trick. Also, you do have to be concerned with getting stalemated, but that's only happened to me once (even though I probably should have been stopped a few times).

Italy: 19 Lepantos (I've never tried anything else). 11 wins, ~58%. Get all of Turkey, turn on Austria in Tri, Gre, and Bul, push toward St. P. and/or Mar/Spa/Por... hope France doesn't come at you at all (F/T is annoying to deal with) and hedge just a little against Austria having A Tri and moving A Vie->Tyo early on if you have to.

France: 9 solos, not quite 50%. Looking through my games I see a lot more early attacks by Italy that I would have expected - it looks like in 6 of the 10 non-solo games, Italy played something aggressive before the end of 1902. In only one of those games was I able to able to fend off Italy while not also getting attacked by E or G or both. Italy also attacked France once in 1905. France attacked Italy in 1902 and got attacked by E/G. (Two of the 10 were F/R/T draws.) Otherwise, when Italy doesn't get involved, it seems pretty easy to roll. France is tied with Turkey with being completely wiped out the fewest times and solos+draws. I still think France is pretty strong.

Germany: 7 Solos, ~37%. Germany seems to be my "feast or famine" country. There were only 2 of the 19 games where I finished with more than 2 but less than 18 centers... Looking back through the failures, there are some E/Fs, some E/Rs, and some F/Rs, and even a couple of E/F/Rs. The successes are widely varied... I have not yet done a deep/thorough analysis of these games yet to determine where I might have played better, but I seem to recall that the choices are tougher and more varied than the general strategies I've been using for, say, Turkey or Italy or France. (Do I let Russia into Sweden? Who do I ally with? Will they actually work with me? Who do I attack first?)

England: 5 Solos, ~26%. Again, like Germany, haven't analysed these in depth yet. Taking a quick look, I see that England has survived a few southern power victories (whereas I have never survived a bot win as Germany). In general, it looks like France's strength is England's weakness... none of the English solos involved a F/E. There were times when I would have liked to do an F/E, but haven't figured out how to convince France of that - France seems to be very inclined to go for Lvp eventually, even if you can stop him from getting it once and he backs off. It appears that E/G is much more workable.

Russia: 5 Solos, ~26%. Very similar results to England. The only thing the successes seem to have in common is being allowed into Sweden in 1901. In those 5 games, only 1 had an R/T alliance that survived until the end of game. In the other 4 solos, Turkey attacked R first in two of them. A couple of the wins has Russia going after Turkey only when Italy had started a successful Lepanto, then kicking Italy out. In the other 14 non-solos, Turkey played Smy->Con->Bul, Ank->BLA in 9, Con->Bul, Smy->Arm, Ank->BLA in 4, and Con->Bul, Ank->Con, Smy->Ank in 1. Of the set of 9, Turkey attacked Russia first in 6 of them, Russia attacked first in 2 of them, and neither attacked the other in one. Turkey seems to like his 1902 attacks, it looks like there were three of those. Russia also had issues with German incursions and dealing with England as well.

Austria: Ick - only 3 solos, a paltry 15.8%. If the board breaks correctly, it's possible - the easiest start would be where Italy attacks Turkey and Turkey attacks Russia. I have not figured out yet (as I mentioned in another post recently) how to get R/T reliably on side (I know I've at least tried a few times but probably not every time) whilst fending off Italy... if Italy and either Turkey or Russia even remotely touches you - they don't even have to take anything, just cut a support (which they seem to do more often than not) - then it's all over.

[DougJoe]

Another update, because I'm in the mood again - I've completed 30 games as every power for a total of 210 games.

Turkey: 21 W (70%), 2 D, 5 S, 2 E
Italy: 17 W (56.67%), 1 D, 6 S, 6 E
France: 14 W (46.67%), 7 D, 5 S, 4 E
Russia: 12 W (40%), 4 D, 6 S, 8 E
Germany: 10 W (33.33%), 6 D, 4 S, 10 E
England: 7 W (23.33%), 5 D, 7 S, 9 E
Austria: 4 W (13.79%), 5 D, 4 S, 15 E

Since my last update 11 rounds ago, Russia's the only one that has seemed to have had a major change, with 7 wins in the last 11 games (63%!). Not sure why that is yet, haven't analyzed the games. Turkey continues to climb, it'll be interesting to see where it levels out. Everybody else seems to be pretty close to where they were before (with rounding):
T: 63%->70%
I: 58%->57%
F: 47%->47%
R: 26%->40%
G: 37%->33%
E: 26%->23%
A:16%->14%

Altogether, this is 85/210, which puts me at a 40.48% solo rate.

If anyone else has numeric data on how they're doing against the bots in classic mode, I'd love to see it, especially if you've played a lot of games.

[Fianwa]

Turkey is very easy for draws, Germany (if you play it right) and Italy are best for wins

[Nescio]

Thanks for sharing your results!
Here's how often I've won so far when playing against bots:

Code: Select all

England :  1 / 11 = 0.09
France  :  1 /  5 = 0.2
Italy   :  2 /  6 = 0.33
Germany :  3 /  5 = 0.6
Austria :  5 / 17 = 0.29
Turkey  :  4 /  4 = 1
Russia  :  0 /  5 = 0
  all   : 16 / 53 = 0.3

I realize my sample is rather small and probably not very representative. Moreover, against bots I tend to experiment with different openings (e.g. Mos→Liv, StP→Bot, Sev→Rum, War→Sil as Russia). Furthermore, I occassionally mess up things by not checking my orders.

As for the bots, are some better than others? I've seen them successfully perform a convoy Tun→Syr as Italy, which is fairly predictable, but I don't recall seeing them block it as Turkey.

[Ogion]

Question: do you all play all the way through, even if you're clearly done for? I understand this is good practice to do that.

[DougJoe]

Thanks for sharing your results!
Here's how often I've won so far when playing against bots:

Code: Select all

England :  1 / 11 = 0.09
France  :  1 /  5 = 0.2
Italy   :  2 /  6 = 0.33
Germany :  3 /  5 = 0.6
Austria :  5 / 17 = 0.29
Turkey  :  4 /  4 = 1
Russia  :  0 /  5 = 0
  all   : 16 / 53 = 0.3

I realize my sample is rather small and probably not very representative. Moreover, against bots I tend to experiment with different openings (e.g. Mos→Liv, StP→Bot, Sev→Rum, War→Sil as Russia). Furthermore, I occassionally mess up things by not checking my orders.

As for the bots, are some better than others? I've seen them successfully perform a convoy Tun→Syr as Italy, which is fairly predictable, but I don't recall seeing them block it as Turkey.

Nice, thanks for sharing as well! 5 out of 17 as Austria seems pretty good!
Apparently my Germany needs work, because everyone else I've read seems to think Germany is easy.

Also, as to which bots are "better" than the others, it's hard to tell because the bots stop playing once the human is eliminated (which skews the stats). The bots seem to do best with Turkey, France, and Germany, at least according to the numbers that I have in my 210 games. (Again, my stats at https://docs.google.com/spreadsheets/d/ ... sp=sharing ).

Question: do you all play all the way through, even if you're clearly done for? I understand this is good practice to do that.

Personally, I do - mostly because I want my statistics to accurately reflect my level of play... I don't want the "failures" to be hidden or masked, so to speak. Plus, sometimes it's actually fun to try to wiggle out of what seems to be impending doom. I could certainly understand why people wouldn't - I rage quit/resign my fair share of chess games after a blunder... but I've seen enough weird things happen with the bots here (especially the later the game goes) that I give it a go.

I just finished this game ( http://webdiplomacy.net/board.php?gameID=356111 ) which was a very rare occurrence of the A/I bots coordinating rather well (often times they don't get the support into AEG correct between them, or Italy simply tries for EMS and is blocked). By the end of 1907, I was down to 1 center, an army in Smyrna. I thought that the A/I kerfuffle in Venice was going to spark a war between them and save me (although it didn't really at the time) and I got back to 3 by the end of 1909. I made a couple of wrong choices in 1910 and 1911 and that was that... but it was fun while it lasted (and I did survive, but I thought I had a chance for a comeback).

I think though, that comebacks from a super-low number of centers to solos are truly fairly rare... this game ( http://webdiplomacy.net/board.php?gameID=339763 ) I think (there's a lot of games to remember!) is the worst position that I've managed to solo from. Had a great start as England, got whooped by Germany mid-game, and in 1918 went from 6 SC to 2 (Por and Mar) and had to disband 4 units. I decided, that winter to dump everything but F Yor and F Por, knowing that Mar was probably doomed and hoping that Por would survive due to the G/I conflict. I thought that, for the most part, it was over - for sure Germany would turn around to chase the fleet in Yor, and I'd eventually lose Por... but he didn't - he bottled up the MAO - and that let me take Edi back. So I was still at two. Using the fleet in Edi as a starting point, I was able to take back Scandinavia and my home centers from Germany while Italy was kept at bay, and then beating through Italy/Iberia/some of Germany's remains... it was a nail-biter all the way, where the 18th center was a fleet that had snuck past the Ionian and could have been blocked out of Con.

Thanks *again* to everyone who was involved in the creation and implementation of the bots!

[Nescio]

Question: do you all play all the way through, even if you're clearly done for? I understand this is good practice to do that.

“Good practice” as in “good sport” or in “good exercise”?
It depends. I'll always issue orders, of course. The objective of the game is to win by controlling the majority of centres on the map, so that's what I aim to achieve.
However, I won't stubbornly go on only to get defeated. Whenever I believe I no longer have a chance of achieving a solo-victory, I'll vote for a draw. I certainly try to survive as long as possible, because all survivors are included in draws. And if at some point other players (or bots) behave unexpectedly, it might allow me to wiggle out of a tricky situation and recover. Getting a hard-fought victory from a near-defeat is far more satisfying than a couple of easy wins.

[Nescio]

Nice, thanks for sharing as well! 5 out of 17 as Austria seems pretty good! Apparently my Germany needs work, because everyone else I've read seems to think Germany is easy.

Austria and Germany have the advantage they're in the centre of the map and therefore have many ways to expand. Provided not all their neighbours simultaneously attack them at the start of the game, of course.

Also, as to which bots are "better" than the others, it's hard to tell because the bots stop playing once the human is eliminated (which skews the stats).

Yeah, that's quite annoying. Kestas Bot seems to have a 9% win rate, though, whereas the other bots are at 2% or 3%.

The bots seem to do best with Turkey, France, and Germany, at least according to the numbers that I have in my 210 games. (Again, my stats at https://docs.google.com/spreadsheets/d/ ... sp=sharing ).

It's nice you keep track of your games. However, I'm not really that interested in which countries survived or how many centres they had.
Although I like winning, I dislike repetitiveness, and I play against bots to try out different things. I prefer Austria to Turkey, which is boring: the latter has only four realistic opening moves, the former dozens; here are the ones I've tried so far:

Code: Select all

Austria (5 / 18)
Rum, Alb, Bud:
  6-way draw, Autumn 1907
Rum, Alb, Gal:
  eliminated, Spring 1904
Rum, Alb, Tri:
  won, Autumn 1908
Rum, Alb, Tyr:
  7-way draw, Autumn 1901
Rum, Ven, Gal:
  5-way draw, Autumn 1907
Ser, Alb, Bud:
  eliminated, Autumn 1906
Ser, Alb, Gal:
  won, Autumn 1909
  won, Autumn 1909
  won, Autumn 1910
  4-way draw, Autumn 1910
  5-way draw, Autumn 1909
Ser, Alb, Tri: 
  6-way draw, Autumn 1906
  eliminated, Autumn 1905
Ser, Alb, Tyr:
  won, Autumn 1908
  survived, Autumn 1913
Ser, Ven, Gal: 
  5-way draw, Spring 1912
  7-way draw, Autumn 1905
Tri, Adr, Tyr: 
  6-way draw, Autumn 1905

It's a small sample and to do it properly, each opening ought to be repeated many times.
I realize 210 games is a lot to go through, yet it would be nice if you could record your opening moves. And if it turns out you usually start with the same moves, try something else more often.

[DougJoe]

It's nice you keep track of your games. However, I'm not really that interested in which countries survived or how many centres they had.

When I started playing against the bots I was curious as to the question of "am I a better player than the AI" which led to the various stat tracking (plus I'm just kind of a stat junkie). I was also curious, since the bot games are gunboat, how the various powers would rank when compared to Swordman's gunboat tier list.

Although I like winning, I dislike repetitiveness, and I play against bots to try out different things. I prefer Austria to Turkey, which is boring: the latter has only four realistic opening moves, the former dozens; here are the ones I've tried so far:

Code: Select all

Austria (5 / 18)
Rum, Alb, Bud:
  6-way draw, Autumn 1907
Rum, Alb, Gal:
  eliminated, Spring 1904
Rum, Alb, Tri:
  won, Autumn 1908
Rum, Alb, Tyr:
  7-way draw, Autumn 1901
Rum, Ven, Gal:
  5-way draw, Autumn 1907
Ser, Alb, Bud:
  eliminated, Autumn 1906
Ser, Alb, Gal:
  won, Autumn 1909
  won, Autumn 1909
  won, Autumn 1910
  4-way draw, Autumn 1910
  5-way draw, Autumn 1909
Ser, Alb, Tri: 
  6-way draw, Autumn 1906
  eliminated, Autumn 1905
Ser, Alb, Tyr:
  won, Autumn 1908
  survived, Autumn 1913
Ser, Ven, Gal: 
  5-way draw, Spring 1912
  7-way draw, Autumn 1905
Tri, Adr, Tyr: 
  6-way draw, Autumn 1905

It's a small sample and to do it properly, each opening ought to be repeated many times.

Those are definitely some interesting approaches to Austria! Interesting to see that you've had the most success with the "common" Ser/Alb/Gal. I'd be interested to see the Rum/Alb/Tri solo if you care to provide a link.

I completely understand the notion of wanting to try different things and not get bored as opposed to simply trying to win all the time.

I realize 210 games is a lot to go through, yet it would be nice if you could record your opening moves. And if it turns out you usually start with the same moves, try something else more often.

I've always been interested in having a true "database' of each of my games, including all of the orders. I believe that others that play here have come up with mechanisms to pull the moves from games through scraping the HTML- I could probably do something similar to see exactly how varied my openings with each power have been, although I've got enough other little hobby projects going on that it might be a while before I get to it... I do, at least, have the game ID list to start with.

[Nescio]

Those are definitely some interesting approaches to Austria! Interesting to see that you've had the most success with the "common" Ser/Alb/Gal. I'd be interested to see the Rum/Alb/Tri solo if you care to provide a link.

That I won win the Ser, Alb, Gal opening a few times is at least partially because I've used it more often, most of the other openings I played only once (and there are still some I haven't tried).
Here's the game you asked for: http://webdiplomacy.net/board.php?gameID=355827
I find this one more interesting, though: http://webdiplomacy.net/board.php?gameID=355819

I completely understand the notion of wanting to try different things and not get bored as opposed to simply trying to win all the time.

Actually I do try to win all the time, that's the objective of the game. I just try to win with different openings, rather than doing the same moves over and over again.

I've always been interested in having a true "database' of each of my games, including all of the orders.

All orders is probably a bit much. Besides, they can already be obtained by following the game links.
Recording only your opening moves makes sense to me: all subsequent orders will (or should) be based on what the other players do.

[Nescio]

And I'm now at one hundred games against bots. Here are the results (i.e. solo victories):

Code: Select all

Turkey  :  6 /  10 = 0.6
Italy   :  6 /  12 = 0.5
Germany :  5 /  11 = 0.455
France  :  4 /  11 = 0.364
Austria :  6 /  21 = 0.286
Russia  :  3 /  21 = 0.143
England :  2 /  15 = 0.133
  all   : 32 / 101 = 0.317

Compare this with my results from a few months ago, when I had completed only half as many games against bots:

Code: Select all

Turkey  :  4 /  4 = 1
Germany :  3 /  5 = 0.6
Italy   :  2 /  6 = 0.33
Austria :  5 / 17 = 0.29
France  :  1 /  5 = 0.2
England :  1 / 11 = 0.09
Russia  :  0 /  5 = 0
  all   : 16 / 53 = 0.3

It's still a rather small sample, though. Of the openings I've used, I did most only once or twice, and there are still many, many more I haven't tried yet. Ideally every opening ought to be repeated a few thousand times.

Perhaps one of the developers could analyse and share the results of all bot games by all players on this site? That would give a more representative outcome.

[Bladerunners]

The bots are programmed to recognize human player and will
concentrate moves on the human player. Ex i played england 12 games in a row where all 3 of F/G/R target england. I played next few games as F/G/R and england left alone in each game and did well. Thats not random ...

[Claesar]

The bots are programmed to recognize human player and will
concentrate moves on the human player. Ex i played england 12 games in a row where all 3 of F/G/R target england. I played next few games as F/G/R and england left alone in each game and did well. Thats not random ...

The bots really cannot. They simply do not have access to that data. However, as they're thinking alike they will understand each other more easily.

[DougJoe]

Don't know if people are still into playing the bots, but here's another update - I'm up to 50 games with each power for a total of 350 games. My results:

Code: Select all

Power	Wins	Win %	D	S	E
Austria	9	18.00%	5	9	27
England	14	28.00%	8	15	13
France	23	46.00%	8	9	10
Germany	16	32.00%	9	7	18
Italy	28	56.00%	2	12	8
Russia	20	40.00%	5	11	14
Turkey	35	70.00%	3	9	3
Total	145	41.43%	40	72	93

Since round 30, the last update, the powers have not moved much.
T: 70%->70%
I: 57%->56%
F: 47%->46%
R: 40%->40%
G: 33%->32%
E: 23%->28%
A:14%->18%

I was really hoping to get France up to 50%, Austria up to 20%, and my overall win rate up to 3/7 (roughly 42%) but alas, it was not to be. At some point I will do some deeper analysis based on what my openings were as well as what the bots like to play, but that's going to take some time.

As usual, games and results at: https://docs.google.com/spreadsheets/d/ ... sp=sharing

I ended up building a couple of graphs that show win rate over time - I really like how the one with all the powers split out looks (the graph is below all the data, scroll down to see it.)

[Trigfea63]

That is some very interesting data, thank you.

I don't have any comparable data, I haven't played nearly enough bot games. But your post made me want to share an anecdote on a game I played recently. I was Italy. By the end of 1906, I had lost all my home centers and was down to 2 units and 2 SCs. And yet ...

https://webdiplomacy.net/board.php?game ... #gamePanel

So, yes, it definitely pays to play bot games out all the way to the end!

Also, I am curious if anyone has any insight into Turkey's (and Austria's) behavior/programming that would allow this to happen.

[kestasjk]

That is some very interesting data, thank you.

I don't have any comparable data, I haven't played nearly enough bot games. But your post made me want to share an anecdote on a game I played recently. I was Italy. By the end of 1906, I had lost all my home centers and was down to 2 units and 2 SCs. And yet ...

https://webdiplomacy.net/board.php?game ... #gamePanel

So, yes, it definitely pays to play bot games out all the way to the end!

Also, I am curious if anyone has any insight into Turkey's (and Austria's) behavior/programming that would allow this to happen.

Yeah .. that is very odd, Ill see if paquette wants to take a look

[Jamiet99uk]

That is some very interesting data, thank you.

I don't have any comparable data, I haven't played nearly enough bot games. But your post made me want to share an anecdote on a game I played recently. I was Italy. By the end of 1906, I had lost all my home centers and was down to 2 units and 2 SCs. And yet ...

https://webdiplomacy.net/board.php?game ... #gamePanel

So, yes, it definitely pays to play bot games out all the way to the end!

Also, I am curious if anyone has any insight into Turkey's (and Austria's) behavior/programming that would allow this to happen.

Yeah .. that is very odd, Ill see if paquette wants to take a look

It almost looks as if once the Bot has progressed its "front line", it does not prioritise defending SCs "behind the line".

Or, in Turkey's case particularly in that game, it looks as if the Bot values *home* SCs and *non-home* SCs as being of equal value, so it does not shift units back to protect home SCs as long as it is winning new SCs elsewhere. If this is part of the explanation, then clearly the programming should be adjusted to account for the fact that home SCs are more valuable due to being able to build only there.

[georgefc3]

Here is an interesting game. This is a Turkey/Italy alliance vs the rest of the board.

I got stuck behind a stalemate line. I was at 10 centers and Italy was at 8 (roughly). I played the same moves for about 5 whole seasons. Then Italy decided to go to the MAO and was somehow able to secure Iberia. Amazing that the other bots didn't stop him.

I let Italy grow. He had a fleet in the Ionian for the longest time, blocking my fleet in the Aegean. He lost a unit at one point and took out the fleet in the Ionian.

I assisted him with my two fleets. One fleet made it to the North Sea!

I eventually stabbed him. I finished with 22 centers including all of Scandinavia and all of Italy proper.

Here is the link:

https://webdiplomacy.net/board.php?gameID=426517

Comments?

[Trigfea63]

Well played. It's pretty unusual to see Turkey with a fleet in North Sea and in control of Scandinavia!