Added building-cost including all resoruces as evaluation-context for more sophisticated building-selection and also as a countermeasure to softlocking a build-order by having no ways to obtain certain resources.
For example, if the AI would drop below 5 wood, while having no market-place and no wood-income it will avoid building any buildings that neither allow trading nor produce wood.
Added trading-logic to Nullkiller-AI.
The AI can now identify which resources it is lacking the most and buy them to fix softlocks in their build-order. It can also sell excess resources that it doesn't have a need for.
Before marketplaces could only be built as part of a requirement for other buildings but not on their own when that other building already existed like it is the case in certain campaign-missions.
There's 3 new evaluation-contexts that are now taken into account:
Whether an action is building, whether an action involves sailing and the newly introduced threat.
The value-evaluation of creatures now also takes special resources into account.
No longer treating other AIs differently than players when it comes to how afraid we shall be of them.
The cost of buildings for decision-making now determines missing resources. Available resources are ignored when it comes to how they impact the cost. But missing-resources will heftily impact the assumed price by calculating their market-value. This shall encourage the AI to rather build what it currently can build instead of saving up for something that it lacking the special resources for.
AI is no longer willing to sacrifice more than 25% of their army for any attack except when it has no towns left.
Revamped the priority-tiers of AI decision-making.
Higest priority is conquering enemy towns and killing enemy heroes. However, the AI will no longer try to do so when the target is more than one turn away and protected by a nearby enemy-hero that could kill the one tasked with dealing with the target. Except when they have no towns left. Then they get desperate and try everything.
As a general rule of thumb one could say the AI will prioritize conquest over collecting freebies over investing army to get something that isn't a city. It's a bit more complex than that but this is roughly what can be expected. It will also highly value their own heroes safety during all this.
Startup-behavior was messing with my intended logic. Mostly by getting excess heroes for no real purpose other than that it could.
This wasted a lot of money that could be better invested on subsequent turns.
I removed it and playing-strength actually went up.
The magic-strength of a hero now checks if the hero has a spellbook and at least one combat-spell.
The impact of knowledge and spellpower to the hero's magic-strength is now also depending on it's current and max mana-pool-size as an empty mana-pool does not exactly contribute well to fights.
Replaced every call of getFightingStrength() with getHeroStrength() which uses both the fightingStrength and the (reworked) magicStrength to guess how much stronger a hero-lead army is.
Fixed that army loss was taken into account both for the path and the target-object. In certain cases, like a hero defending a town, this could lead to armyloss being twice as high as it should be.
Heroes with conquest-tasks will only endanger themselves to be killed when they can execute a conquest in the same turn.
Heroes with other tasks will dismiss any tasks except of defending when they'd be within one turn of an enemy hero that could kill them.
Tasks are now in different priority-tiers. For now there's 2 tiers. One for regular tasks and one for tasks of the new "conquest"-type. Regular tasks will only be considered when no possible conquest-type tasks were found.
Slightly reworked scoring heuristics.
Openmap is no longer tied to difficulty-level due to being configurable anyways.
Tasks are now in different priority-tiers. For now there's 2 tiers. One for regular tasks and one for tasks of the new "conquest"-type. Regular tasks will only be considered when no possible conquest-type tasks were found.
Recruit-hero-behavior is now evaluated before movement to make it more likely a new hero can exchange their stuff with others.
No longer excluding paths for exposing a hero to an enemy in the behaviors. There definitely are reasons for doing something anyways, even if threatened. The logic for that should be done in the PriorityEvaluator.
Supressing hiring army on turn one seems just bad. Starting the main-hero as strong as possible seems like a good idea to me and hiring the available troops outright will help achieve that goal.
However, if there's a hero for hire, who has army with him that is a better deal, we hire that one first.
When we have no hero, we will definitely want to hire one.
We will also want to hire heroes who already pay for more than themselves by coming with an army that has more value than the hero costs.
Due to morale-considerations the AI sometimes calculated that their strongest army after doing an exchange had slightly lower total value than the army they used before.
But by using unsigned "slightly lower" became near infinite.
So they constantly wanted to upgrade their army because they considered it more useful than anything else.
Changing the unsigned into signed fixes this.
It is now possible to switch to an AI-variant that uses hand-written heuristics for decision-making rather than the FuzzyLite-engine. This is configurable in nkai-settings.json via the new parameter "useFuzzy".
The information of whether objects like a redwood-observatory or subterranian gates have been interacted with by the AI will now be retrieved from the game-state instead of using an AI-internal memory that won't survive loading a save-game.
Nullkiller suggested that this change would help to further fix inconsistent behavior by the AI. I tested it and it did indeed fix different orders of how AI does things.
"Important to make count 1 to not relay on object addresses
They are source of random" - Nullkiller