pguillory/xathis and combat

[Euphony]

Yay, pguillory reached #1 and passed xathis! Kudos to pguillory. Personally I'm really glad to see xathis' bot, which has held that top position since the very beginning of the contest, finally get passed. I look forward to much heated contest between the top bots for that coveted first place. You guys are fun to watch!

So, what was it that kept xathis in the top spot for so long? I think we all know at this point it was combat, and that's what pguillory has surpassed xathis in. Yeah, what this contest boils down to is combat algorithms. Sure, when we first started out, it was all about path-finding or collaborative diffusion or however we are getting to food, but soon enough we played in a real game and realized that we won't even get to the food unless we have some strong combat implemented.

Where to improve? Well, I noticed that among the very top bots, there usually reaches a kind of stalemate between two forces at the choke points, whether it be in a maze or cell_maze map. A certain amount of ants, sometimes as much as 10+, wall off against each other and can make no progress. Since I know nothing about combat, I throw it to the forum, what can happen here? Is it possible to break these standoffs (say if all ten ants move forward at the same time), or is it an inherent part of the maps?

[Memetix]

The times, they are a changing ....

Having watched quite a few of the top bots playing, I'm not sure it is really about the combat, more about making the best use of the ants you have.
Both Xathis and Pguillory are adept at building defensive walls and not having too many wasted ants behind these walls. Other bots that build up massive blocks of ants are effectively wasting the ants that aren't in the front row, these could be off exploring and gathering food.

There is a downside to being efficient and building a defensive wall 1 deep. An aggressive bot who doesn't mind swapping ants can break though the wall with a mass advance and the area behind the wall is often sparsely populated and a quick land grab is possible is you have some ants ready to move in.

So there seem to be 2 main strategies emerging.
1. Build a wall with the least number of ants required to stop the opponent advancing and then use your other ants to explore and find food, gaining more ants and out growing the opposition.
2. Build up a large army and have some "break through" code to spot when it is worthwhile sacrificing some ants to gain territory. In multi-player games this will probably benefit the players not involved in the combat more than the one breaking through the wall.

I suspect coding option 1 is simpler, but the metagame suggests a good option 2 bot can do well at the moment. If you can come up with a solution that incorporates both options you may just win the competition.

[agent_smith]

About pguillory, I told you so

Walls at choke points are in my opinion a function of bots being generally cautious: neither one goes all in. It might be a good strategy to behave more aggressively if you have ants to spare.

[Euphony]

Memetix,

Xathis released a bot version #2, and the scenario you were describing came up : http://aichallenge.org/visualizer.php?game=211610

After the two other bots were destroyed, it was you vs. xathis. You had almost twice as many ants, but xathis appeared to be the one who tried to break through the choke point. He sacrificed and sacrificed until it was a mop-up job for you.

Xathis' new bot must have thought it was ahead to sacrifice rather than defend. I think this technique is new to version #2. It appears he addressed the issue we were discussing.

Would you like to comment?

[BenJackson]

That may is particularly unfair. Whoever is paired with the weaker opponent has a tremendous advantage. In fact, I use it that way specifically in home testing to artificially boost my own old versions to make them tougher opponents for my new work.

That game was decided by "oldman" and his excessive sacrifices in his defense (where barsandcat did a very creditable job considering the rank of 50).

[McLeopold]

Xathis failed to explore the center section when he could have. Instead he concentrated on fighting the hill on his side. Part of a good strategy is being balanced and gathering information about the entire map.

[Memetix]

I do a lot of my testing on that map .... it provides choke points that need sacrifice code to break through.
It seems that both Xathis(v2) and my bots were both taking the initiative to advance at different times.

Having watched Xathis' old bot, it tended to win in the late game due to territory dominance. It would often come from behind and be the last man standing. The downside was it wasn't as aggressive as some other bots and could be many victory points behind early on, and if it didn't gain the upper hand in the middle game it would often come last. I suspect the new code will result in slightly less 1st place finishes but a lot less last place finishes.

Whenever I watch his bot in action I wish mine was that accurate with its moves and overall strategy.
The behaviour of all my ants is emergent so I don't have any simple way to get them to form defensive lines or form a grid to gather food efficiently in "my territory".

[agent_smith]

It seems to me that in that game, Memetix was able to deal with its local opponent more quickly and so grabbed more territory, after that it was a matter of time.

[Euphony]

Yes, that map (maze_04p_02) is bad, and I assume it won't be used in the final competition. It's the cell_mazes we should be looking at.

However, that game as well as many other of xathis' cell_maze games does illustrate that xathis' bot is much more aggressive than it used to be. The issue addressed as point #2 in Memetix' first post about having 'break-through' code when you have a large enough army seems to be a major component of this top bot. The only thing is, xathis attempts to break through even when he has a small army. Even when he has a smaller army than his opponent. Sometimes he gets pushed back after losses. Yet, he wins, and wins, and wins.

This isn't the best example, but it kinda shows what I'm talking about : http://aichallenge.org/visualizer.php?game=212112
At move 428, xathis manages to take the ant hill with 3 ants, even though pguillory was there first and has five ants. Also, at move 577 at the fight at row 58 col 156. If you'll watch up to the end of the game, you'll see xathis is sacrificing even with a smaller army. These situations between pguillory and xathis *used* to be stalemates. Now, there is more aggressive code. The improvement has happened. I am satisfied.

aggressiveness + excellent line formation with ants = speed at taking enemy ant hills

PS congrats on results against xathis, Memetix