Topic: Training sites

Woot! I did win nile map in less than 10 hours. 9:53:25, to be exact. It wasn't easy. Oh, beating the AI was. I lost 9 soldiers in the whole game, and I always had power superiority in every front. but the map was so big, and my soldiers taking long time to refill military sites, i thought i wasn't going to make it in time.

So, let's see what I can say:

On some aspect AI has improved. On some it is worse. Difficult to judge exactly because I myself have improved a lot from the time I played with AI, but I noticed AI started to use the training sites after about two hours. before, it took like 4 hours. So, progress. On the other hand, even with plenty of time to develop, AI had only a very small number of trained soldier. most of them didn't even have the evasion. Old AI was fairly weak, but if you gave it 4 or 5 hours it would get the evade promotion on every soldier it had. So, one step forward and one step back. And I see that now AI dismantles buildingss that are disconnected from economy. that meant that once I cut away some branch of enemy land (happened often for how nile map is made) enemy would dismantle everything, leading to its soldiers dieing because they were too far away from warehouses to reach them. Although in the end that slowed me down, it took me more to make my own bildings than to conquer enemy ones. Also, I see that empire still never makes the arena. no imperial soldier that i saw was promoted to evade 2.

I will be glad to see the result of your game

I take it means that you would like to receive the replay by mail. I'll send it as soon as i finish writing.

I am aware that AI does not make differencies based on map size. F.e. in estimation how many wood it needs (it has internal targets) or it could adjust a time when first trainingsite is to be build.

One thing I would do that is adjusted for mapsize is how many foresters and woodcutters one makes in the beginning. the more you have to grow, the more wood you need, while in a smaller map making plenty of wood would take away precious space from the rest of the economy. So, while in desert tournament I made 4 foresters (6 with atalntis for smokeries) in nile I made 8 foresters and woodcutters before I did anything else, and I soon doubled them. AI should also have similar parameters, linking the map size to the amount of foresters it will try to make early. Possible problem: if the map is plenty of water-mountains-lava, the AI may make too many. only fertile terrain counts for that. eventually a command could be added to dismantle the excess ones after expansion is ended.

It is theoretically possible to cluster productionsites, in fact something like this is done with space consumers, even though it might not be that much visible. If we identified some building as belonging to industry X, AI could scan vicinity of possible building field, and increase score if buildings of the same industries are found.

Yeah, I think that would be a good compromise. Especially important for foresters because they work better when together.

re cheating by AI - It would be more elegant if we allowed various amount of starting wares for AI - if we had easy/medium/difficult game levels. But generally I am against cheating here....

I think the best way to chet with AI would be to make its buildings faster. So the time needed to accomplish a task, instead of being T, would be T*D, where D is a parameter of the difficulty, and it would be 1 for medium level. We could set an easy level (D=1.25), hard (D=0.8) very hard (D=0.6) and godlike (D=0.4). That would keep the gaming experience similar at all difficulties level; it would not be good if more difficulty meant that the AI has a starting spike, but then becomes useless later. My idea would speed up the whole AI economy, in a manner consistent in all phases of the game. Notice that woodcutters, foresters, fishers, would not work as fast as everything else, because their waiting time is reduced, but the time they take to reach a spot and do their job will remain the same. but they will still be faster overall. So at godlike the farms woud produce food 2.5 times faster, and give it to the mines which would consume it 2.5 times faster, making metal that will be used by the metal economy also 2.5 times faster... I think that would be really challenging. I'm not sure I could beat it in 1v7.

After all, every strategic or 4x game has difficulty levels where AI has increased production. It's not cheating, it's balancing for the fact that it does not have a brain and is not good at handling "soft" variables.

Top

Quote

Tibor wrote:

king_of_nowhere wrote:

On some aspect AI has improved. On some it is worse.

Please be more specific, you are comparing what (revision/branch) with what (revision/branch)

With what I understand he is comparing Build 18 and 7428 version (or which is current official dev build?)

einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/
backup website files: http://kartezjusz.ddns.net/upload/widelands/

Top

Quote

From my experience:

king_of_nowhere wrote:

I see that AI now makes ships, but I suppose it uses it only for transport.

They are sending one ship for expedition, then they are going round and round with that ship. Till good spot is found. If none spot left- going rond forever.

colonizing is too situation-dependend to be well done by ai.

Yes it is. But most of maps with seafaring needs colonizing.

Empire players have been the more consistently able to get good promotions for soldiers (...) because of the differences in the economy, or just an accident of statistics.

Probably because their economy is very simple to get lvl 10 soldiers, plus you have creating level 0 soldiers much harder than barbarians. Atleantans need upgrading 2 stats in each building. So in both building you need special wares. Other tribes don't need much efford to get level 2 soldiers, fully upgraded in one building. One building is always easier to manage comparing to 2 buildings

einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/
backup website files: http://kartezjusz.ddns.net/upload/widelands/

Top

Quote

einstein13 wrote:

They are sending one ship for expedition, then they are going round and round with that ship. Till good spot is found. If none spot left- going rond forever.

I need to fix this (set some timeout on that last expedition), but on good map it has couple of colonies before all suitalbe spots are exhausted and that never-ending expedition takes place.

Top

Quote

sincce this thread has gone off topic and become about AI improvement, I think I'll continue here... today I played in the tournament, and at a point i was disconnected. the oothers quitted and rehosted from the saved game, but in that saved game (we used one of theirs) the AI had gotten control of my land for a few seconds. And it managed to screw me up pretty big in such a short time. It dismantled all my castles away from the border. Some of those were on the border with my ally, and they were keeping my ground, so the moment they were dismantled I also lost ground and a few farms got destroied.

Luckily that had no impact on the game, which was already won, but it showcased another weakness of the AI. It should be programmed to keep the border with its allies, not dismantle everything and lose ground. Also, keeping some back fortresses is a useful strategy, if your front ccollapses it pays to have additional lines to fall back to. and it especially pays on the AI, since it always loses (I mean, even if it was vastly improved, humans generally set a difficulty level such that they win. I never saw a human who liked to lose with AI, only difference is between those who set to win hard and those who like to win easy) it would make things more interesting for it to keep a deeper military line.

And since we are in argument of AI limits, in version 18 AI did not build scouts, and if it could not see an enemy fortress it would never be able to attack and would get stuck. has that been improved, or is it still possible to keep an AI at bay forever just by building a castle in a clever place?

Top

Quote