HOI4 Dev Diary - South Africa - together for victory?

[Tirenedon]

It is more the other way around - South Africa goes communist if they go down the communist side of the tree (although it is certainly worthwhile to consider tinkering a bit more with the AI weights to account for outside interference through boosting party popularity).

So does that mean that if I as the Soviet Union made South Africa go Communist after they started down the Allies side of the tree they will continue to go down that side of the tree and help the Allies even if they also join the Comintern? I suppose the same could be said about any other country.

 

[Axe99]

Programming AI weights to provide an optimal choice for every possible situation is a task that can consume a nearly endless amount of time. For now we are mostly focused on making sure the AI is not quite as eager to ally itself to a Germany that is currently occupied by Soviet troops.

This be a good thing to hear .

In terms of avoiding a database-driven strategy, how about adding a slight (to medium) random weight to the weights (through an extra factor, so it's not in everything, and there's scripting control over how random it is) - that'd nip any "just 0.5 per cent more influence and [insert thing here] gameplay", and it wasn't as if decisions around international relations (or pretty much anything) were able to be calculated precisely in any event. As long as the random factor isn't too large, it shouldn't lead to crazy (well, crazier, depending on the situation ) results. Just a random thought, ignore if not helpful .

 

[Alex_brunius]

This be a good thing to hear .

In terms of avoiding a database-driven strategy, how about adding a slight (to medium) random weight to the weights (through an extra factor, so it's not in everything, and there's scripting control over how random it is) - that'd nip any "just 0.5 per cent more influence and [insert thing here] gameplay", and it wasn't as if decisions around international relations (or pretty much anything) were able to be calculated precisely in any event. As long as the random factor isn't too large, it shouldn't lead to crazy (well, crazier, depending on the situation ) results. Just a random thought, ignore if not helpful .

This randomization already happen with for example AIs scripted research choices ( the AI often don't pick the choice with highest weight ), and as far as I know also with the diplomatic AI in how it responds to some events.

As many modders are already familiar with you can use the console command aiview to for example see what research priorities the AI would have.

 

[Axe99]

This randomization already happen with for example AIs scripted research choices ( the AI often don't pick the choice with highest weight ), and as far as I know also with the diplomatic AI in how it responds to some events.

As many modders are already familiar with you can use the console command aiview to for example see what research priorities the AI would have.

There's definitely a randomisation effect with research choices (and a define to specify how random it is), but we're not talking about research here . Do we know this relates to national focus, idea choice or other weights (I'm dopey atm, can't think any other weights off my head, but think they're there) though? Even if it does, it's an engine-applied "one weight fits all" approach, instead of enabling scripters the potential to decide how random something is (which I think is fine for research, but less useful for NF and idea choices).

Edit: Using AIview (cheers for the tip, hadn't played around with that before), you get some pretty fractional weights in the tech tree (up to three decimal places), but with NFs, it's in whole integers, which may suggest the tech tree randomisation isn't applied to the NF tree at least.

 

[Shock360]

There's definitely a randomisation effect with research choices (and a define to specify how random it is), but we're not talking about research here . Do we know this relates to national focus, idea choice or other weights (I'm dopey atm, can't think any other weights off my head, but think they're there) though? Even if it does, it's an engine-applied "one weight fits all" approach, instead of enabling scripters the potential to decide how random something is (which I think is fine for research, but less useful for NF and idea choices).

Edit: Using AIview (cheers for the tip, hadn't played around with that before), you get some pretty fractional weights in the tech tree (up to three decimal places), but with NFs, it's in whole integers, which may suggest the tech tree randomisation isn't applied to the NF tree at least.

It's still applied, but with focus tree's there's less options so they don't have to have as in depth percentages. However one big problem with this is there doesn't seem to be enough factors- there needs to be a massive overhaul of the requirements and ai factors do that they make sense

 

[Lukevan16]

@podcat
I hate to sound demanding, but will you guys be considering the minor geographical changes I suggested earlier? I just didn't know if you'd seen it

 

[Axe99]

It's still applied, but with focus tree's there's less options so they don't have to have as in depth percentages. However one big problem with this is there doesn't seem to be enough factors- there needs to be a massive overhaul of the requirements and ai factors do that they make sense

Do we know this for sure? We know the weights have random numbers applied against them (or the AI would just choose the larger weight each time), but do we know the weights themselves are altered?

Regardless, my main point probably relates more to triggers for events (as there's no way it would be possible to know an exact 'break-point' for an NF or research decision in any case). I'm replying to an issue mentioned by a dev, so I'm assuming if there was a random factor involved in the context I'm talking about, that it would have made it a non-issue (as a random factor means it's not possible to know the break point).

 

[Shock360]

Do we know this for sure? We know the weights have random numbers applied against them (or the AI would just choose the larger weight each time), but do we know the weights themselves are altered?

Regardless, my main point probably relates more to triggers for events (as there's no way it would be possible to know an exact 'break-point' for an NF or research decision in any case). I'm replying to an issue mentioned by a dev, so I'm assuming if there was a random factor involved in the context I'm talking about, that it would have made it a non-issue (as a random factor means it's not possible to know the break point).

What do you mean the weights are altered? As in no modifiers change the chance of them being chosen, or the actual base is never changed?

 

[Axe99]

What do you mean the weights are altered? As in no modifiers change the chance of them being chosen, or the actual base is never changed?

@Alex_brunius noted that for research there was a random factor beyond the weights (which we know there is), such that the next tech chosen by the AI is chosen effectively based on AI_weight + random factor (so, in effect, the random factor alters the base weight), so it's not just a simple probability choice based on weight. So, in this context, do national focus choices involve a similar AI weight plus random factor, such that say you had two NF options, one with a weight of 9 and one with a weight of 1 - is it a 90%/10% chance, or is there an introduced random factor that means its 90 +/- X% / 10 +/- X% chance, where X% is the random factor (and if it's added to one side, it's removed from the other - and +/- could be multiplicative as well - I obviously don't know whether it's applied or not, let alone the nature of the operator involved).

 

[Chimeraguard]

New member here, just got HoI IV as my first Paradox game(aside from a brief stint with Stellaris) a couple of weeks ago, and have been having a lot of fun with it so far.

Looking forward to seeing Together for Victory and all the fun stuff to do as the Commonwealth. Will there be a new set of achievements to go with the DLC?

Also, I wonder what fun stuff the Raj will have on its side. Can't wait.

 

[Cody MacArthur Fett]

It is more the other way around - South Africa goes communist if they go down the communist side of the tree (although it is certainly worthwhile to consider tinkering a bit more with the AI weights to account for outside interference through boosting party popularity).

So it's impossible in these new foci trees for the AI to change ideologies unless they go down certain branches? That's probably not how it works, but I want to be clear instead of just assuming.

The theoretical scenario I was imagining for that question was as such though: Player is controlling Soviet Union (in Ironman, so no tagging), USSR boosts Communist influence in South Africa, and SA turns into Cape Commune thanks to USSR meddling before deciding on the political branch of the foci tree. Will they then go down the Communist branch, or will they still go down the Commonwealth branch?

To add a variation into that scenario, suppose that SA has already chosen Support the Policy of Appeasement and maybe some of the foci after. Will the new Cape Commune get the chance to go down the Communist branch now, or will the Allies have a Communist nation among their number, or will the CC be up a creek without an Alliance or foci tree?

The relations with Soviet Union thing is an interesting idea. We've been wanting to make relations between countries more relevant, but there are a few issues with that - mostly the fact that it forces people to dive into the files to find the exact breakpoint at which the AI becomes more likely to do a thing for them. We call that "database-driven strategy", and it is something we have tried to avoid.

I can definitely see that. One of HOI4's greatest strengths is that it is clean, not overwhelming people with a fantastically large amount of data or the need to dig through the files. However, at the same time I am a player who has started to realize that relationship value is really only good when positive for an extra 10 points to the desire to accept a deal.

I'm not a programmer though, so all I can do is wish you and the rest of the talented Paradox team luck on solving it.

Programming AI weights to provide an optimal choice for every possible situation is a task that can consume a nearly endless amount of time. For now we are mostly focused on making sure the AI is not quite as eager to ally itself to a Germany that is currently occupied by Soviet troops.

That is a very good priority list.

Thank you for the prompt and informative answers, sir.

 

[Alex_brunius]

Edit: Using AIview (cheers for the tip, hadn't played around with that before), you get some pretty fractional weights in the tech tree (up to three decimal places), but with NFs, it's in whole integers, which may suggest the tech tree randomisation isn't applied to the NF tree at least.

Does it even matter when the AI will pursue exact order of NFs when you have "historical focuses" ticked which I assume is the normal case here?

 

[Archangel85]

So does that mean that if I as the Soviet Union made South Africa go Communist after they started down the Allies side of the tree they will continue to go down that side of the tree and help the Allies even if they also join the Comintern? I suppose the same could be said about any other country.

Well, you can't be in two factions at once. But yes, once a country starts down a mutually exclusive tree branch, they are locked in. We don't currently have the option to reset NF trees or even just branches (it is quite a complicated issue as we would have to remove all boni and account for the exploitation potential of first getting all the boni from one branch, and then resetting the tree to get all the boni from the other). But the AI generally prefers to finish industry-boosting foci first (since it is nearly always the most optimal choice), so you do have a bit of time before they make a choice.

 

[rcbricker33]

For the diplo weights could you not put in adjusting modifiers based on relations. So if SA is forced to communist, that doesn't automatically push them towards the soviets (although I think it is realistic that it improves their opinion of the USSR) but it could change the weight so that it becomes more likely that they snuggle up to Stalin. Even if the USSR were to be the one to boost and force the change. This could have a change in weight modifiers for a number of things, like Lend/Lease or Volunteers. You could also do this with if X country attacks Y country. Depending on the AI relationship with those countries the Modifier is adjusted accordingly.

SO if I play Bolivia and attack Peru, Ecuador's weight would be adjusted by modifiers for my WT addition, My relationship with them, their relationship with Peru, My Aggression on the same continent with them...etc. Might make them more likely to join my faction (or want to) if my relationship with them is high enough, or to join a faction against me.

just a thought to stop data-base strategies and make the AI behave a little more realistically.

 

[Teutonizer]

Eh, I almost always read DDs such as this a few days later... (not a good thing when you're really like the game)
Still, hope the TfV will be ready soon enough, coz I can't even play when I'm expecting new cool features to be added (and also, can't wait to play as South Africa)

 

[Axe99]

Does it even matter when the AI will pursue exact order of NFs when you have "historical focuses" ticked which I assume is the normal case here?

Aye, weights are definitely irrelevant in historical mode. I wasn't really focused on the national focuses - as best I can guess, the kind of things that would need to be monitoring for particular levels of relations or the like would be event triggers (the NF triggers are shown in the NF trees, so there'd be no need to go data diving to find the triggers in any event), and as far as I can tell (noting I haven't looked deeply into it - I'm still very much learning the modding side of things) for the triggers beyond the MTTH there aren't a lot of randomised elements (although by all means school me on this if I've missed something ).

 

[Gamengervi]

Well, you can't be in two factions at once. But yes, once a country starts down a mutually exclusive tree branch, they are locked in. We don't currently have the option to reset NF trees or even just branches (it is quite a complicated issue as we would have to remove all boni and account for the exploitation potential of first getting all the boni from one branch, and then resetting the tree to get all the boni from the other). But the AI generally prefers to finish industry-boosting foci first (since it is nearly always the most optimal choice), so you do have a bit of time before they make a choice.

Why is your post background not dark like the other devs?

 

[Archangel85]

Why is your post background not dark like the other devs?

Because I'm not good at remembering to set stuff to be posted as PDS rather than a regular user.

 

[Gamengervi]

Because I'm not good at remembering to set stuff to be posted as PDS rather than a regular user.

 

[terminalstatus]

Very cool