AI combat discussion

[kyrub]

Very interesting discussion.

ILS, your third picture is really correct.

Invaders 1 and 2 can attack the gate guardian "a". "b" can't be attacked directly.

1. So, invader 3 cannot attack A?? This looks like a bug, either the invader 1 should not be able to attack A, or all invaders 1-3 should be able to do so - what do you think?

2. So, nobody can attack B? I think I saw some different situation where it was possible... but I'm not sure.

3. I will list in as a bug the fact that the defender at the gate is not fully protected. This seems simply wrong, gates gave special, massive protection to the defenders.

[I like Serena]

1. So, invader 3 cannot attack A?? This looks like a bug, either the invader 1 should not be able to attack A, or all invaders 1-3 should be able to do so - what do you think?

Sorry, bad copy + paste. I fixed it in the previous post to read "1, 2 and 3".

Edit: I also added another remark about the defense bonus if the defender attacks.

--ILSe

[b0rsuk]

Ah, this makes more sense than you know.
Looking at the code, I think the defender gets his shield bonus only in defense!
So if he attacks, the City Walls play no part. (Can you confirm that?)

I can't confirm it, but I suspected it. I avoid attacking with the gate guardian, but attack with gate guardian's neighbor. And I try to have basilisk/cavalry nearby. It's weird how cavalry can charge through a wall, but as long as the mechanic is not changed the best I can do is to give (some) halberdiers the same ability.

+1 is a bit low, yes.

About the city walls layout: I was trying to represent things that can't be represented. Walls are not another square type, but rather a border between squares. Towers are impassable. It would be interesting if towers could be entered from inside and provided the usual +3 bonus (and prevented melee attacks).

About letting a single invader move in and ganging up on him: I suppose this can be effective, but 1) bonus for gate guardian sounds more intuitive, a naive guess would work 2) buildings get destroyed as soon as something passes the gate, so you'd have to code a special case.

[b0rsuk]

Magical ranged (hero) attacks
The AI should use magical hero attacks much more often. Currently it has strong preference towards spells. At 3 mana per shot, it's a great deal compared to many spells. Ideally, the chance to use a ranged attack would be proportional to ranged strength of a hero.

Casting spells in nodes
In node battles, the AI should have strong preference for casting spells matching the color of the node. In absence of such spells, it should casts spells matching the color of a mastery (sorcery, chaos, nature) it has. I haven't checked the code, but based on my experience the 3 masteries make it twice as hard to counter the spell, not to dispel it in M:tg sense.

[kyrub]

(noted you suggestions, b0rsuk)

Time for some good news, we have more straight bugs here, which means potentially more fun when they are gone:

AI casts too many defensive spells in the mid-late game. I hope I am not alone who thinks this is a problem in the current game. Defensive spells are not particularly game-winning in current model and AI seemed to love them. I tried to change their priority in last patches, but it was not still the right blend. Now, I see there is a bug: AI mistakenly gives too much weight to spells (like 5 times the normal one!) that can negate some effect you have on your units. Correcting this should have a massive effect - we will now hopefully see more spell alternatives in battles and how the AI uses them! - It won't be much more clever, but more diverse certainly.

Thanks to ILSe, I have found out that there is a whole AI feature that does not work: counter spells priority. Helas, it is tied to defensive spells so it is less exciting... Still, at least the AI will use Magic Immunity, Counter spell, Righteousness more in situations where you actually have some mana points to cast.

[b0rsuk]

More effective gate guarding
The AI doesn't care which units end up guarding the gate. It should put the most durable units in there (flyers, if you want AI to be really mean). Then, if the gate is already secure, it should try to place a hard hitter in the spot next to the gate. Ideally a creature with first strike or breath attack. Also it would be nice if AI would save healing spells (from priests) for the gate guardian (Especially if it has ranged troops).

Saving ammo for better opportunities
Generally, there are cases where the AI would benefit from waiting until enemies are "in range". Especially when AI is defending a city. A common tactic for dealing with AI cities is to wait until it's out of ammo, then move in. However, the gate makes a good bottleneck, you just can't kill the defending archers quickly even if you're close. In most cases, it will be safe for AI to wait until enemies are closer. (It's not safe to wait when the attacker has strong ranged capabilities or spells to shut down archers).

[fantasma]

The AI will always cast spells, even if it flees the same turn.

A simple exploit is repeatedly initiating some fight where the AI has little chance to win or overwhelming advantage but you can get away with little losses.

It will spend all his casting skill, so you can drain his reserve. So, before storming a well defended city, circle around and attacking the wandering stacks for a while can be very useful. This is old experience from vanilla, so maybe it has been fixed already.

And I second b0rsuk's suggestion/observation. If I am facing ranged attacks I move my troops back to the closest hill until he runs out of ammo. Then I close in. The AI could/should wait until you close or chase you in many situations.

Is there a possibility to set some optimal range which it tries to maintain, i.e. just out of melee range?

[kyrub]

It will spend all his casting skill, so you can drain his reserve.

Thanks for that one, fantasma. This problem is much deeper than being just one exploitable issue. The game often ends by you exhausting AI in terms of mana spent, once you start to fight many battles a turn (not sure if it is still the problem in 1.40h but I bet it is).

The AI is taught to cast spells every turn, without no exception, until it runs out of skill available. Thus, when its wizard skill gets higher, it also represents a bigger danger for the AI: it will spend 2-3*skill per battle, wearing itself down.

I must say that I tried to find a solution for this, but I still did come with a suitable algorithm. My last conviction is that AI should somehow set itself to spend only X% of skill at the beginning of the battle (but how should I determine that X? based on total mana? based on battles fought this turn? still not good enough...).

By the way, AI casting something damaging before fleeing may be a good idea, especially if that something is Cracks call on your hero. So, disallowing this = dumbing down AI. The set of conditions must be more sophisticated.

[Catwalk]

I think this is actually handled adequately in the game, by giving the AI players really large mana reserves most of the game. It becomes increasingly difficult to drain them like this.

[luddite]

Yeah I feel like any change to that AI behavior would hurt the AI more than help it. On hard/impossible level, the AI pretty much never runs out of mana, and their biggest strength is their ability to cast multiple full-strength lightning bolts at the beginning of every battle.