AI

[Catwalk]

I meant whenever you can block its approach so it can't go around the spearman.

Another abusive strategy: Move your top defenders out of your city and keep only spearmen inside it. Voila, no strong stack will ever approach you. Would that work? I agree that the only solution is probably to refrain from using it. Alternately, I'd love it if attacking a stack didn't end your turn. The AI should be able to use that just as well as the human player.

[Seravy]

I meant whenever you can block its approach so it can't go around the spearman.

Another abusive strategy: Move your top defenders out of your city and keep only spearmen inside it. Voila, no strong stack will ever approach you. Would that work?

Probably not. "Value" is determined by buildings and population in case of cities. Military power is used for "value" only if nothing else is present except units. (Nodes have a value according to their aura size, etc)
Unless that stack is really powerful in which case killing it is probably a larger blow than taking the city, the city will get attacked.
If you mean it'll get targeted by a weak stack first, possible, but that's pretty much random. Whichever stack contains the oldest unit on the continent will pick the target first and attack the city, and there is no guarantee that will be a weak stack. And if the good stack is closer than 4 tiles (or you have other cities to attack), it won't work.

Unfortunately if the spearmen is blocking, there is nothing else that can be done. Waiting and not attacking is worse - it effectively wastes the stack without even using it.

[Catwalk]

What I would do as a human player is to use a few units on a tiny stack, not my entire stack of doom. Does the AI know how to split up stacks? Or does it only do that when it needs to garrison something?

[Seravy]

What I would do as a human player is to use a few units on a tiny stack, not my entire stack of doom. Does the AI know how to split up stacks? Or does it only do that when it needs to garrison something?

No it doesn't. I'm happy it can finally build them, splitting one up after it was built would be horrible. - it isn't as simple as the unit moving back to rejoin next turn, the AI cannot merge stacks like that. It can only designate a rally point - one per continent - and gather units there from the entire continent, selecting the 9 strongest available for that.
Garrison needs can pull units into cities  but only from stacks that are still idle and doing so is last in the line, so if the big stack is moving towards any target, it won't get split this way.

[Seravy]

I'm wondering how many ships and/or floating islands the AI should try to have.

Currently it uses the formula, 6+(Turns/32)*(5-Land Size) for ships, and a fixed "3" for floating islands - existing ships included so if the AI has 3 ships, it won't make islands.

(where land size is 0 for Tiny and 4 for Huge)

However, there seem to be two problems with this.

1. While larger land means naval travel is less important for expansion, larger land size means more productive cities, resulting in more troops to transport (and more enemies to defeat) so more ships might be needed for effective invasions - ultimately, if the AI has only 9 ships on huge, it means it can only transport 9 stacks to the destination continent in the time period it takes for a ship to travel between the shores of the two continents - easily a dozen turns if the enemy is not adjacent. While that's still plenty for most difficulty levels, it might not be for the highest ones?
2. If the AI knows Floating Island, it won't prioritize Ship Wrights Guild early, and might not build it at all for a long while - resulting in being limited to the 3 floating islands for all transporting needs. This amount is clearly insufficient to support the new "many settlers on huge impossible Myrran" AI tactic - land bound settlers made on already filled continents will get stuck waiting for a free island for many turns, especially as they have to compete with troops requiring transport.

So I think the ships might be fine as is, but Floating Islands should have either the same formula as ships (can be a drastical waste of casting skill if the human player finds and kills them) or the early ship wright's guild should be enforced on high difficulty Myrran AI even if they already own Floating Islands (in which case they currently do not prioritize it at all, and even without the spell, the priority bonus on Huge for SWG is not high), or maybe the number of islands could be bumped up to 5 on impossible to support the faster expansion (and more early units and especially summons)?

...or we can call it a non-issue as this recent test shows the Myrran AI to grow twice as fast as any others despite having only 3 floating islands and zero ships - 3 settlers still waiting to get transported? (It built 30 cities in 100 turns which is about twice as good as it used to be before this change)

[Nelphine]

Probably a non issue. Also probably reasonable to allow 4 or 5 islands on extreme and 6-8 on impossible. They'll have plenty if casting skill for it.

[Catwalk]

What I would do as a human player is to use a few units on a tiny stack, not my entire stack of doom. Does the AI know how to split up stacks? Or does it only do that when it needs to garrison something?

No it doesn't. I'm happy it can finally build them, splitting one up after it was built would be horrible. - it isn't as simple as the unit moving back to rejoin next turn, the AI cannot merge stacks like that. It can only designate a rally point - one per continent - and gather units there from the entire continent, selecting the 9 strongest available for that.
Garrison needs can pull units into cities  but only from stacks that are still idle and doing so is last in the line, so if the big stack is moving towards any target, it won't get split this way.

In my nomad game I saw a stack of 9 trolls and other riff raff take a node, leave 3 behind and move on with the other 6. What determines this behaviour?

[Nelphine]

Nodes can also build garrisons similar to a city, as can towers.

[Seravy]

What I would do as a human player is to use a few units on a tiny stack, not my entire stack of doom. Does the AI know how to split up stacks? Or does it only do that when it needs to garrison something?

No it doesn't. I'm happy it can finally build them, splitting one up after it was built would be horrible. - it isn't as simple as the unit moving back to rejoin next turn, the AI cannot merge stacks like that. It can only designate a rally point - one per continent - and gather units there from the entire continent, selecting the 9 strongest available for that.
Garrison needs can pull units into cities  but only from stacks that are still idle and doing so is last in the line, so if the big stack is moving towards any target, it won't get split this way.

In my nomad game I saw a stack of 9 trolls and other riff raff take a node, leave 3 behind and move on with the other 6. What determines this behaviour?

Nodes have their "required minimal garrison size to keep" set to 3. So any units in excess of that can be pulled into the stackbuilding point, if they are the best units on the continent. However, nodes also have a "required garrison amount" of 9, so they also pull unused units back to fill it up. This mechanism allows the AI to replace strong units with weaker ones and avoid wasting their best heroes/sky drakes on garrisoning a node for the entire game.

[Catwalk]

Does the AI favour casting spells of the same colour as the node when fighting in a node? I just fought Raven in a sorcery node, I know he has Psionic Blast but he chose to cast (and fail) Call Centaurs instead.