|
Post by gp1628 on Mar 7, 2012 12:23:47 GMT -5
OK I see this a lot but so far cant pin it down. If someone has some specifics then please start a new thread. But here is mine.
I was playing High Cultist the AI was playing Baron at Knight level the map was LARGE
At first I did think there was something wrong. Often I saw large armies near my positions. They seemed obviously larger than mine but did not attack. Especially in sites that I thought the Baron was particularly fond of.
Baron 143 vs HC 33 in a Town (1 shield) he attacked
Baron 171 vs HC 71 in a Castle (3 shields) he did not attack
Baron 23 vs HC 6 at a Mine (1 shield) did attack
Baron 47 vs HC 23 in a Town (1 shield) did attack
Baron 12 vs HC 4 on a mountain (1 shield) did attack
Baron 83 vs 45 on plains did attack
So these would seem to be confusing. But it seems that the Baron AI (remember that it is specific to the Baron player) takes those shields into account and whether or it has knights in its group. I no longer easy access to the hard numbers that the AI uses to compare armies. But loosely it seems that if I have 10 on the plains then they count as 10. If its on a mountain or in a town then that 1 shield doubles it to 20. And in a castle the 3 shields make it times 3? or 4? So the Baron wants to outnumber you PLUS your shield bonuses and still have enough army left after the fight to leave some behind or have an army to move on with. And in the case of the Baron, knights seem to count extra.
|
|
|
Post by gp1628 on Mar 7, 2012 12:30:01 GMT -5
Im going to try and watch the debug file more. I think that the actual comparison numbers might be recorded there.
Altho to do a really GOOD test I would probably have to use a loaded map. I might be able to guess at which battle in the file is the one Im looking at. But Im pretty sure the debug log records them as X Y locations. If I can have that map also loaded into the map editor at the same time then I will be able to visually get the X Y of each battle position.
|
|
|
Post by gp1628 on Mar 7, 2012 12:39:45 GMT -5
OK as a partial judgement: The Baron AI is not playing as aggressively as I would play it. But then that isnt necessarily bad. The times that I do win while playing as the Baron its not because I judged my attacks better than he does. It tends to be more from making good use of luck or map reading which would be a subject for a separate thread.
IF the AI did play the Baron like I do then he would often attack just based on "I have more units than you". He would continue doing so marching outward in a straight line until his army was whittled down to nothing, and then start a new army out from the Citadel. Sometimes that works for me but Im not sure if it would be good for the AI to do.
Maybe this is all perception of the AIs level. What if we ask that the Baron AI play that aggressively on the LOWER AI settings like Piss Boy and Jester. After all, Jester is supposed to be "human normal starting player" level of AI. If it played that aggressively instead of spreadsheet smart, then I bet we would get less complaints about the AI being "stupid". And it would make the higher level AIs seem smarter because they would be playing more carefully as people tend to expect from a higher AI setting.
|
|
|
Post by gp1628 on Mar 7, 2012 12:44:34 GMT -5
There is also Map size to figure in. On the smaller maps, the Baron AI (and some others) should play more aggressively. Less carefully.
|
|
|
Post by gp1628 on May 31, 2012 10:45:40 GMT -5
OK I keep testing the things being said that "everyone knows" and the way they are worded they all tend to always be wrong. I find VERY little that the AI ALWAYS does. I have had the AI target me. But the Baron and Senator tend to target my capitals. The Witch and Druid tend to target my armies. Those make sense. So I suspect when someone says " the AI wont hunt me down like it should" they are seeing a lack in a nation which is doing what it is meant to be doing. The AI doesnt attack when it obviously has the advantage. I have had the AI stomp me with armies, and attack with multiple armies. I think many players do not understand the weight that the AI gives certain units and especially the shields of a terrain (mountain, tower, castle). But what I see is that the AI seems to want an advantage of twice as many points as I have. Its hard to be sure since we no longer have the i key to see the army strength numbers it uses. Again, Im not sure what Id recommend. The actions seem logical. And they work much better at higher AI settings than they do at lower AI settings so it makes it seem that the AI does things different (better) at higher settings. Thats good. The actions are logical, but not human enough? Should we give up some of the "intelligence" of AI to get AH? (almost human). Add some random in so it can sometimes be stupid? The complaints seem to be grouped with "the AI is too careless early in the game" and "the AI is too careful late in the game". Can we work that into the AI? Can the AI play careful early in the game? And then switch from wanting twice as much army to wanting 1.5 as much army when it seems to have a strong lead? Im not sure how to measure that. If its based on being ahead on the scoreboard, then that can happen on turn 5 when the AI has almost no army. My math sucks. Can someone come up with an AI formula that takes in turn # and score and my army size versus your army size (which I think automatically includes defense bonuses like being on plains versus being in the castle). Then we can ask Johan to use it on a random 10% chance? That might give us an AI that seems to sometimes take-a-chance like humans do.
|
|
|
Post by ulius on Jun 3, 2012 7:14:25 GMT -5
You certainly put an emphasis on the strong points of the AI and I don't think it's bad on determining which stacks to attack (but it still fails against Creeping Doom right now. I was attacked recently by a stack that had no magic damage) There are still some important things that could be added: - does the AI have a way to maneuver its stacks tactically and strategically? It seems to be true that the AI attacks with multiple stacks, when they are near. But has it a way to maneuver them in the vicinity of commander/citadel it wants to attack? Does it move them where the player might go?
- Is it actively trying to boost squads in order to defeat certain stacks, who are moving near their captital, for example?
- Is the AI able to compose powerful stacks? (I've seen many Necromancers running around with longdeads and soulless, but it would still hire spearman instead of crossbows/archers)
- Can it counter certain units? When it sees a Creeping Doom, will it give a very high priority to hire wizards? Will it maneuver its wizards to the Creeping Doom?
- Is it able to analyze the worth of hunting down an independent? It often seems to leave indies standing next to its income sites alive and has to come back later to capture them again (and again .... and again ... and again...)
- Is it able to maneuver around threats? A big problem with it now is that the AI doesn't recognize "Stacks of Doom". It will send one of it's smaller stacks after the other to certain death, even if its united stacks would be powerful enough to win.
- Is it able to react to threats? Does it react to a stack in the vicinity of its last citadel? Does it try to save its last commander?
- Does it actively try to citadel-kill a player (and I don't mean randomly)?
- Does it burn down ancient forests/troll forests when playing against TK?
I think many complaints about the AI are about different questions, than the one you answered. It seems good enough to me, when it comes down to the decision of "attack - yes or no?". When complaining about lack of aggression, the main problem is imho, that the AI has many stacks running around somewhere and it will not try to maneuver them to a worthy target (like the last commander of a player). It has seven times of your army size, but only a part of that will be near the player and the AI doesn't try to change that fact. It won't save its smaller stacks from annihilation and it won't build/unite stacks in order to defeat the stacks it has been facing. edit: I'd like to add another example of a tactic the AI is unable to adapt to: Play with Barbarian against a high lvl AI. Try to survive and focus on herb income. Summon as many spirits as you can and use the commanders for herb collection. As soon as the AI gets close to you (and it will be a lot stronger than you mathematically) do the following: Swarm around the borders of the AI with your ghost commanders and try to capture its income sites with your floating ghosts. After capturing run away. You don't need to run far, two squares is usually enough, as long as you don't stand in a line to the next income site you captured. After you identified where the AIs stacks are push further with your ghosts that are safe and retreat them after conquering sites deeper in the borders of your enemy. Retake sites two turns after the AI takes them back to give your ghosts enough time to run away again. Keep on pushing back the borders of the AI until it becomes impossible to further do this (because its empire is too small). Now you have a lot more income than the AI and can build up your armies very relaxed. I was surprised at how few ghost commanders I lose when doing something like this. Actually most of the ghosts I lose die, because an independent attacks them. Furthermore this will keep the AI from attacking you in any way (at least that's what happens in my games). It will withdraw units that are near you in order to take back sites (even ones far away), that it has lost. If you keep it busy with enough ghost commanders you can have almost no army at all. In this example you can see, that the AI obviously isn't aggressive enough. It doesn't try to kill the weak ghost armies even, if they are very close, because it doesn't see them. It doesn't split up its armies to defend its sites. It doesn't attack the player, although it is a lot stronger than him and has the advantage. And this hasn't to do much with its ability to determine the strenght of a stack, so it won't help to fine-tune that. I want to underline, that aggressiveness is a more complicated construct, and users complaining about "lack of aggressiveness" might mean something very different from what you and I think it means.
|
|
|
Post by gp1628 on Jun 3, 2012 11:17:20 GMT -5
Many good points. Thank you for someone finally responding For now it appears that goal setting of the AI is a lot like the movement settings allowed in modding. We cried loudly for individual AIs (in the forums ofr Illwinters previous game) and seemed to have gotten it. But so far its individual on "who I am" and not very active on "who I am fighting". So whether the AI acts against Citadels is more based on whether or not its a nation that wants citadels (Baron yes, Druid no). Effectiveness against raiding tactics such as you describe would have a lot to do with whether its a nation that wanders to collect everything vs one that goes more directly to its goals (Druid good, Baron bad) I agree that it would be great to write in specific changes of tactics based on the opponents tactics. If we want to start listing them, I will gladly pass them on. That would be something for someone far more organized than I to come up with. Im more of a lover of random. Its not as good but its easier to do. Something between "always do the smart thing" (too predictable and abusable) vs the other extreme of always doing a totally random thing (stupid happens far more often than a surprisingly good tactic). So my idea for now is to come up with good general tactics, some wide variables for when it should be used, and a reasonable number for the random chance of it being used. If you want something more specific than we probably need a new thread for that.
|
|
|
Post by gp1628 on Jun 3, 2012 11:17:45 GMT -5
does the AI have a way to maneuver its stacks tactically and strategically? It seems to be true that the AI attacks with multiple stacks, when they are near. But has it a way to maneuver them in the vicinity of commander/citadel it wants to attack? Does it move them where the player might go? At the moment the AI seems willing to march clear across the map to get to its goal. But only certain nations seem to set their goal on citadels, others on commanders. So multiple stack attacks seem to be more luck from having multiple stacks with the same goal. And it happens more often with higher AI settings (I know its a side-effect of the purchasing boosts but it still amounts to higher AIs playing better).
So for this situation, my easy-ass suggestion would be that every nation should have a percentage chance to flip once in awhile. From citadel to commander hunting, and vice versa.
I dont know that a recognition of whether a nation they are up against would be better beat by selectively changing targets could be formulized easily. Altho come to think of it, a suggestion of primary/secondary targets might help. So Baron would by "Capitals or Commanders if nearby" and Druid would be the reverse.
ByTheWay the AIs do seem to place a higher rating on home capitals. Both in taking any, and in retaking any it finds.
Is it actively trying to boost squads in order to defeat certain stacks, who are moving near their captital, for example? That is something that could be addressed. The AI seems to concentrate its stack building on "front line" locations. So it doesnt concentrate on backline sites, even its capital, unless it gets taken.
Is the AI able to compose powerful stacks? (I've seen many Necromancers running around with longdeads and soulless, but it would still hire spearman instead of crossbows/archers) A very good point. I will post that now. It fits in with the "each having separate AI" benefit we have. Certain AIs (such as Necromancers) need to have a higher priority setting on purchasing Archers. Come to think of it, I think that applies to almost every class. The general setting should be higher.
Can it counter certain units? When it sees a Creeping Doom, will it give a very high priority to hire wizards? Will it maneuver its wizards to the Creeping Doom? On the one hand, I want to answer NO. But on the other hand, I think they all prioritize purchasing mages whenever offered, and tend to use them as commanders to send out squads. So Im not sure what change we would want. "IF Creeping Doom, then dont send stack until 3 mages" instead of sending it out with one?
Is it able to analyze the worth of hunting down an independent? It often seems to leave indies standing next to its income sites alive and has to come back later to capture them again (and again .... and again ... and again...) Again, that would be depending on the class I think. Some do and some dont, and its based on their theme. Its not that they "analyze" anything. Its just that some are more commander hunting than others. That would fit my "lets have a random flip of goals" suggestion which I figure would be easier than analyzing
Is it able to maneuver around threats? A big problem with it now is that the AI doesn't recognize "Stacks of Doom". It will send one of it's smaller stacks after the other to certain death, even if its united stacks would be powerful enough to win. It does? I thought it was overly careful about wanting big numbers vs small numbers before attacking. But it does seem to not assign proper weights to some enemy units.
Is it able to react to threats? Does it react to a stack in the vicinity of its last citadel? Does it try to save its last commander? It does lack recognition of standings. Its strategy does not seem to take into account "I am high scoreboard" or "I am low scoreboard". Nor whether the other guy is high or low scoreboard. (scoreboard being recognition of army, capital, resources, etc)
Does it actively try to citadel-kill a player (and I don't mean randomly)? Some do and some dont but it appears to be based more on "who am I" than on "who are they". This could be helped by recognition such as "if they are Witch then dont chase their stacks, chase their capitals". But that changes the classes themes. And with modded nations starting to show up, it wouldnt work well. Maybe make it general such as "if he has lots of army and low on capital numbers then target capitals" but at higher settings that almost happens accidentally. Again I think an easy thing would be randomly changing focus once in awhile so that if he is low on either one it will occassionally get targeted.
Does it burn down ancient forests/troll forests when playing against TK? More by "who am I" than "who are they". Certain nations have a tendency to burn forests in general. And others really hate the idea in general. Illwinters famous for doing balance "rock/paper/scissors" more than "one-vs-one"
|
|