Improve AI alliance logic #992

OptimizedForDensity · 2024-01-20T17:53:37Z

Mainly reworks AI alliance logic but also does some other clean up and implements a few missing pieces of logic regarding diplomacy actions changing relations. The alliance logic is based off of Victoria 2 to the best that I could recreate, which means changing the sequence of boolean checks with a running score which means alliances can be a bit more flexible. There are a few places where it wasn't obvious how the game was calculating everything but I tried to get as close to the original as possible.

The logic is shown in ai_will_accept_alliance in a pretty straightforward manner and I did try to be mindful of performance by adding in an early exit that can reduce the number of calculations. Overall the AI tends to have more stable alliances, and things like spherelings accepting alliances from other GPs, countries with negative relations forming alliances, and countries accepting alliances from nations who own their cores (except for vassal/substates) should be fixed which makes the gameplay more consistent with V2. The GP alliance limit is now active so nations can have only 1 GP ally before Great Wars are invented then they can have 2 (which is what I believe is the case in V2). The AI are now also required to have diplomatic points available for the make/cancel alliance action.

The prune alliance logic was also redone to be consistent with the form alliance logic since I found that the AI was too capricious with making then breaking alliances.

I haven't tested this in multiplayer so I don't know if you'd want other devs to do testing with that before putting this into a release.

Add diplomacy function Final pass

schombert · 2024-01-20T21:07:37Z

src/nations/nations.cpp

@@ -906,15 +936,29 @@ std::vector<dcon::political_party_id> get_active_political_parties(sys::state& s
 	return parties;
 }

-void monthly_adjust_relationship(sys::state& state, dcon::nation_id a, dcon::nation_id b, float delta) {
+dcon::diplomatic_relation_id get_diplomatic_relation(sys::state& state, dcon::nation_id a, dcon::nation_id b, bool force_create) {


The way that dcon works is that it is safe to look up values from the invalid handle, and such values will always be zero. Thus, when there was no relationship object between nations, looking up the relationship would return the invalid handle, and then looking up the relationship value from that would return 0. This helped cut down on the number of relationship objects. If we created a relationship object every time we checked the relations between two nations we would quickly head towards n squared such objects.

I do see that you have force create as an extra parameter, so I realize that this won't actually be creating such relationships everywhere. It just makes my danger sense tingle. So my comment above isn't a blocking objection.

schombert · 2024-01-20T21:10:06Z

src/nations/nations.cpp

+}
+
+// also returns true if one is the sphere leader of the other
+bool in_same_sphere(sys::state& state, dcon::nation_id a, dcon::nation_id b) {


this can be simplified to just return a_sphere == b || b_sphere == a || a_sphere == b_sphere && a_sphere; under the assumption that we only pass valid handles

schombert · 2024-01-20T21:12:38Z

src/nations/nations.cpp

+}
+
+bool are_neighbors(sys::state& state, dcon::nation_id a, dcon::nation_id b) {
+	for(auto g : fatten(state.world, a).get_nation_adjacency()) {


This can be reduced to get_nation_adjacency_by_nation_adjacency_pair. This does a (fast) hash lookup using the combined indices and will return a valid handle (true if cast to bool) if they are adjacent and an invalid handle if they are not

schombert · 2024-01-20T21:21:27Z

src/ai/ai.cpp

+bool has_our_core(sys::state& state, dcon::nation_id n, dcon::nation_id other) {
+	for(auto prov : state.world.nation_get_province_ownership(other)) {
+		auto fat = fatten(state.world, prov).get_province();
+		for(auto c : fat.get_core_as_province()) {


get_core_by_prov_tag_key does a hash map lookup for existing cores using the province and national identity id combination, which could replace this inner loop. However, it may be even faster to loop over all the cores of n's identity and check if any of those is owned by other.

schombert · 2024-01-20T21:26:07Z

src/ai/ai.cpp

+	uint32_t upper_gp_ally_limit = (state.military_definitions.great_wars_enabled) ? uint32_t(2) : uint32_t(1);
+	uint32_t gp_ally_count = uint32_t(0);
+
+	std::vector<dcon::nation_id> allies;


I would rather not have what feels like an unnecessary vector doing allocations here. Just looping over the diplomatic relationships and checking for gp allies directly would be more efficient. If you want to factor out duplicate code from that process, you can write an iterator that iterates over just the allies of a nation, allowing you to write for(auto a : alliesof(state, from)) { ... } which then allows you to even break out early if you hit the upper_gp_ally_limit

schombert · 2024-01-20T21:26:21Z

src/ai/ai.cpp

+	uint32_t gp_ally_count = uint32_t(0);
+
+	std::vector<dcon::nation_id> allies;
+	nations::get_allies(state, n, allies);


schombert · 2024-01-20T21:27:08Z

src/ai/ai.cpp

+float number_of_alliances(sys::state& state, dcon::nation_id target, dcon::nation_id from) {
+	uint32_t ally_count = 0;
+	std::vector<dcon::nation_id> allies;
+	nations::get_allies(state, target, allies);


schombert · 2024-01-20T21:28:29Z

src/ai/ai.cpp

+float current_wars(sys::state& state, dcon::nation_id target, dcon::nation_id from) {
+	float value = 0.0f;
+	std::vector<dcon::nation_id> allies;
+	nations::get_allies(state, target, allies);


schombert · 2024-01-20T21:30:08Z

src/ai/ai.cpp

+		}
+	}
+
+	if(military::are_in_common_war(state, from, target) && !military::are_at_war(state, from, target)) {


you may want military::are_allied_in_war here -- that is true if they are on the same side of some ongoing war

schombert · 2024-01-20T21:32:08Z

src/ai/ai.cpp

+			continue;
+		if((n.get_overlord_as_subject().get_ruler()))
+			continue;
+		if(state.world.nation_get_diplomatic_points(n) < state.defines.alliance_diplomatic_cost)


generally, we let AIs ignore diplomatic point costs. There has even been some talk of removing diplomatic points from the game entirely

schombert · 2024-01-20T21:35:43Z

src/ai/ai.cpp

-				} else {
-					break;
-				}
+		for(auto other : prune_targets) {


It seems to me that this will make the AI drop allies without considering its overall strategic situation, causing it to drop allies it may very well need to deter an attack

The main reason for this change was that I felt that this condition differed from the alliance forming logic, as the AI would make allies without considering their strongest neighbor's strength explicitly but then it could end alliances based on the strongest neighbor's strength which meant that a nation could meet the criteria for both. The hope was that the combination of distance penalty and strength differential would work as a crude strategic assessment, but I might play around with tweaking the new logic so the code I removed would be partially reincorporated (maybe giving a bonus/penalty based on how the new/current ally fares compared to the strongest neighbor).

Alright, I am just concerned that, now that alliance formation is affected by "feelings," a nation could get mad at its strong ally, break the alliance, and then immediately get jumped. It is always very disappointing when this happens because it tends to stand out (you ally with X; an event makes X mad at you; X breaks the alliance; X immediately gets wrecked)

schombert · 2024-01-20T21:36:50Z

src/ai/ai.cpp

 		return false;

 	// Has not surpassed infamy limit
 	if(state.world.nation_get_infamy(target) >= state.defines.badboy_limit * 0.75f)
 		return false;

+	// vassals always ally liege


vassals don't ally their liege in PA because they are automatically called into wars with their overlord regardless of alliance status. It is thus easier to leave them as not allied

schombert · 2024-01-20T21:38:43Z

src/ai/ai.cpp

 bool ai_will_accept_alliance(sys::state& state, dcon::nation_id target, dcon::nation_id from) {
-	if(!state.world.nation_get_ai_is_threatened(target))


why have you decided to remove this check? If the AI is not threatened, it means that it does not think that it needs more allies to deter an attack. At that point, additional allies provide no material benefit while limiting its expansion options.

schombert · 2024-01-20T21:43:26Z

src/ai/ai.cpp

+				else
+					return a.index() > b.index();
+			});
+			command::execute_ask_for_alliance(state, n, alliance_targets[0]);


so, assuming that you have excluded human-controlled nations from the list of possibilities, the list of possibilities should only include valid targets that would accept the alliance. Thus you can go directly to make_alliance and skip the redundant checks. The reason we always exclude human-controlled nations from the list of potential targets is to not annoy the human player. The human player is assumed to proactively ask for alliances that he or she wants, and if the human player isn't asking, then they are assumed not to want one. Hence, we don't annoy them with asks that are expected to be rejected.

schombert · 2024-01-29T23:52:56Z

Is work on this PR still ongoing?

OptimizedForDensity · 2024-01-30T22:45:55Z

Yes and I have a bunch of your feedback already addressed, but I probably won't get around to another big push until this weekend.

OptimizedForDensity · 2025-01-18T18:54:29Z

but I probably won't get around to another big push until this weekend

It's been a very long weekend but obviously too many things have changed and I can't even get the release build to work on Linux so I don't plan to continue this. I based most of this alliance logic on Victoria 2's values in the diplomacy screen, but between not keeping up with the game for a year and new non-V2 gameplay being added, I'm not sure what all is necessary/needed. I'll keep this open in case someone wants to pick it up, but if it's not needed then go ahead and close it.

Improve AI alliance logic

ddc45f9

Add diplomacy function Final pass

OptimizedForDensity force-pushed the AllianceLogic branch from c19ad3b to ddc45f9 Compare January 20, 2024 17:58

schombert reviewed Jan 20, 2024

View reviewed changes

OptimizedForDensity marked this pull request as draft January 30, 2024 22:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve AI alliance logic #992

Improve AI alliance logic #992

OptimizedForDensity commented Jan 20, 2024 •

edited

Loading

schombert Jan 20, 2024

schombert Jan 20, 2024 •

edited

Loading

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

OptimizedForDensity Jan 21, 2024

schombert Jan 21, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert Jan 20, 2024

schombert commented Jan 29, 2024

OptimizedForDensity commented Jan 30, 2024

OptimizedForDensity commented Jan 18, 2025

		bool ai_will_accept_alliance(sys::state& state, dcon::nation_id target, dcon::nation_id from) {
		if(!state.world.nation_get_ai_is_threatened(target))

Improve AI alliance logic #992

Are you sure you want to change the base?

Improve AI alliance logic #992

Conversation

OptimizedForDensity commented Jan 20, 2024 • edited Loading

Choose a reason for hiding this comment

schombert Jan 20, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

schombert commented Jan 29, 2024

OptimizedForDensity commented Jan 30, 2024

OptimizedForDensity commented Jan 18, 2025

OptimizedForDensity commented Jan 20, 2024 •

edited

Loading

schombert Jan 20, 2024 •

edited

Loading