HOI4 Dev Diary - New Zealand & Combat Log

[Ruck]

I dont wanna be unpolite, but I have to say that Im really absolutely disapointed of the whole DLC and Patch 1.3.

The game as it was released was so long good enough, besides an extreme lack of AI abilities (In my personal opinion I call it "not really working")
Even on hardest difficulty mode and all sliders to the most right, you dont have to make more than just building enough 7:2 Inf:Art divisions, you dont need tanks and you also dont need 0 Airforce, just build 7:2 divisions with Poland, draw a frontline, click go and conquer germany and italy with about 60-70 of that divisions, after that draw a second frontline on Soviet Russia (now ca. 100 divs), Speed 5 and wait, correct frontline for Krim and the region north of Leningrad after some time, thats it, now you have conqured all of Europe and Soviet Asia and that on the hardest difficulty modes/sliders.

This didnt changed since the first day of release.
But now I can buy a DLC where I can play an extended New Zealand ? Really ? This is a game about WW2..... There are so many interesting minor countries like Hungary, Romania, Switzerland, Sweden, even Mexiko could be interesting, but New Zealand ? Im not talking that NZ is in common not interesting, but for WW2 ?

Things which me personally interests for DLC-Patch

- Totally reworked AI < = this should be in any case a free patch like it is now, but I really would like to see it earlier, why do we have to wait for the release of the DLC?
- Reworked Airforce = more comfort options
- Better Lategame Performance ( Something must be bugged in the code, as if you conquer most countries, the overall units on the world are much less than in the beginning, but the performance is anyways worse )

What i like is the new Blitzmode ( So I can directly say in which provinces exactly my troops shall go right ?)

 

[Alex_brunius]

How can you be disappointed with a DLC and patch before having played them?

There are so many interesting minor countries like Hungary, Romania, Switzerland, Sweden, even Mexiko could be interesting, but New Zealand ? Im not talking that NZ is in common not interesting, but for WW2 ?

Neither Sweden, Mexico nor Switzerland had any serious part in WW2 historically, while New Zealand did. That makes them alot more interesting for WW2 in my opinion.

In total around 140000 from New Zealand served overseas during WW2, so they did certainly contribute to the war.

 

[jodirola]

oh baby look that atrittion <3 like EUIV..... extremely ridiculous!

 

[KiwiNoob]

However, I'd note that you're currently being out-languaged by an Australian

Thats when you know you've hit rock bottom

I'm not quite sure what you mean by combat AI though. Do you mean the tactics chosen by commanders, or the attacker to defender ratio when choosing to attack, or the choice of the province to attack? There's not much AI (at all) involved in the actual combat (beyond naval combat, where there is actually a bit going on - and I fully agree, I'd love more control over it, as there's definitely room to improve it). Apologies if I'm missing something really obvious, and you're going "How thick is this Australian?!"

Sorry - it was a little bit vague. By the combat AI/battle AI I mean the part of the AI that decides how to organize divisions along the front, what attacks to make and from where, what defensive battles need to be reinforced and how to shuffle units around to account for the changing front, how to exploit weaknesses in the front and counter attack, etc... The combat/battle AI is where the most pressing need is.

Followed very closely by the global army management part of the AI which determines how many and what type of divisions should be assigned to which theater/front line, how many and what type of divisions should be assigned to port defense and suppression and how to properly create new fronts as needed (after a successful naval invasion). At the moment the AI choices are:

1. Do nothing and let a small number of weak divisions take your entire country (*looks at Italy*)
2. Abandon all active fronts and ship 200 divisions in to deal with the 8 enemy divisions that just landed.

Hope that makes more sense

 

[KiwiNoob]

@KiwiNoob It is safest for me just to quote a Dev on this subject of AI development:

And that post points to another problem - their 'fixes' are not always well thought out or tested. Removing strat redeployment along a front line was a good example of a 'fix' that made things worse not better.

Lets also deconstruct the quote you mentioned:

"For example, we implemented a change that the AI should not consider a border with a country it has a NAP as something that needs to be defended. This makes sense, obviously, since that's the entire point of a NAP."​

No - this doesn't make sense. Simply removing the border all together from the AI consideration makes no sense at all. The AI should still consider it but just take into account the NAP. If it has a NAP and it's fresh or the other country has no armies there then sure - you can leave it unguarded. If the NAP could be broken or the other country starts amassing troops along the border then yes - you should start thinking about putting your own ones there too.

"But it also means that the Soviet AI doesn't defend against Germany if the Molotov-Ribbentrop Pact happens. So we removed that"​

A) If you tell the AI to ignore something why be surprised when it ignores it?
B) Another all-or-nothing. Instead of implementing a proper fix they just switched it off again.

"But then Germany considered the long border with the Soviet Union to be important enough to stick a couple dozen divisions there. That in turn led to the German AI having trouble beating France. So we reverted that change."​

They've even identified the exact issue here that the German AI considers the Soviet border too important. The "obvious" solution here to adjust the AI to consider it slightly less important. But no - instead: "So we reverted that change"

Sigh.

 

[Axe99]

Sorry - it was a little bit vague. By the combat AI/battle AI I mean the part of the AI that decides how to organize divisions along the front, what attacks to make and from where, what defensive battles need to be reinforced and how to shuffle units around to account for the changing front, how to exploit weaknesses in the front and counter attack, etc... The combat/battle AI is where the most pressing need is.

Followed very closely by the global army management part of the AI which determines how many and what type of divisions should be assigned to which theater/front line, how many and what type of divisions should be assigned to port defense and suppression and how to properly create new fronts as needed (after a successful naval invasion). At the moment the AI choices are:

1. Do nothing and let a small number of weak divisions take your entire country (*looks at Italy*)
2. Abandon all active fronts and ship 200 divisions in to deal with the 8 enemy divisions that just landed.

Hope that makes more sense

That makes a lot more sense . I'd probably call the first the operational level AI (noting that I'm hardly an authority on these things), and the second the strategic level (in terms of deployments, not builds). I'd also agree that they're the two areas in need of the most improvement . That said, I'd be very surprised if we don't see some improvement in that area (but by all means rub in your superior cross-Tasman predictive performance if it does turn out that way ).

 

[tom_jones]

They've even identified the exact issue here that the German AI considers the Soviet border too important. The "obvious" solution here to adjust the AI to consider it slightly less important. But no - instead: "So we reverted that change"

I don't think it's quite that simple, or obvious -- making the AI consider a border "slightly less important" sounds fine in void, but what's the actual "correct" behavior this evaluation is supposed to produce?

E.g.: Germany has ongoing war with France and the Low Countries, and let's say these countries combined have total number of divisions comparable to Germany. And then Germany has also long border with the USSR (which is currently under NAP) that it needs to consider as "slightly less important". Now, how many divisions is Germany supposed to pull from the western front, to man that slightly less important, potential eastern front -- a dozen? A couple of dozens? More than that? Let's not beat around the bush, no matter what the AI will in the end decide to do, there will be a group of people who will complain that the AI is dumb, the only difference will be that it's dumb because #1 it left eastern front unmanned, or #2 it's dumb because it only placed token force on the eastern front, or #3 it's dumb because it pulled decent number of divisions to man the eastern front, but now it's too weak on the western front and cannot make fast enough progress there.

(and then for extra points we can consider situation where the USSR actually starts moving more divisions to their border with Germany, and so how exactly is the AI expected to deal with this development, keeping in mind it still has ongoing war with enemies who are far from being a pushover..)

 

[Riktol]

I don't think it's quite that simple, or obvious -- making the AI consider a border "slightly less important" sounds fine in void, but what's the actual "correct" behavior this evaluation is supposed to produce?

E.g.: Germany has ongoing war with France and the Low Countries, and let's say these countries combined have total number of divisions comparable to Germany. And then Germany has also long border with the USSR (which is currently under NAP) that it needs to consider as "slightly less important". Now, how many divisions is Germany supposed to pull from the western front, to man that slightly less important, potential eastern front -- a dozen? A couple of dozens? More than that? Let's not beat around the bush, no matter what the AI will in the end decide to do, there will be a group of people who will complain that the AI is dumb, the only difference will be that it's dumb because #1 it left eastern front unmanned, or #2 it's dumb because it only placed token force on the eastern front, or #3 it's dumb because it pulled decent number of divisions to man the eastern front, but now it's too weak on the western front and cannot make fast enough progress there.

(and then for extra points we can consider situation where the USSR actually starts moving more divisions to their border with Germany, and so how exactly is the AI expected to deal with this development, keeping in mind it still has ongoing war with enemies who are far from being a pushover..)

Basically in that scenario AI Germany needs to have won earlier. We should give the AI the ability to rage-quit or, more usefully, time travel.
More realistically, look at what a human player* does and attempt to model that.
I'm not 100% sure but I think the answer is a combination of micro and armoured schwerpunkt, neither of which are available to the AI.

*might be worth specifying a successful human player

 

[Wraith11B]

@podcat will the logs be able to be exported as a text/spreadsheet (or even just set up to be) so that AAR people don't have to write a book to get an AAR out?

 

[KiwiNoob]

I don't think it's quite that simple, or obvious -- making the AI consider a border "slightly less important" sounds fine in void, but what's the actual "correct" behavior this evaluation is supposed to produce?

E.g.: Germany has ongoing war with France and the Low Countries, and let's say these countries combined have total number of divisions comparable to Germany. And then Germany has also long border with the USSR (which is currently under NAP) that it needs to consider as "slightly less important". Now, how many divisions is Germany supposed to pull from the western front, to man that slightly less important, potential eastern front -- a dozen? A couple of dozens? More than that? Let's not beat around the bush, no matter what the AI will in the end decide to do, there will be a group of people who will complain that the AI is dumb, the only difference will be that it's dumb because #1 it left eastern front unmanned, or #2 it's dumb because it only placed token force on the eastern front, or #3 it's dumb because it pulled decent number of divisions to man the eastern front, but now it's too weak on the western front and cannot make fast enough progress there.

(and then for extra points we can consider situation where the USSR actually starts moving more divisions to their border with Germany, and so how exactly is the AI expected to deal with this development, keeping in mind it still has ongoing war with enemies who are far from being a pushover..)

You are 100% right that it is not quite that simple, otherwise poor AI wouldn't be something that plagues many games. But it's also not quite that complicated. Getting the simple decisions right (E.g. do I attack across the large river into mountains or across no river into plains?) would see 80% of the problems gone.

You're post asks some good questions though and it's the sort of scrutiny that the AI needs to have applied to it. The answer would definitely depend on how important the Western front is compared to the Eastern front. If pulling troops away from France would result in a stalemate or loss then better to move no troops out east. Likewise if the AI determines that it can spare 10 divisions but that it would need 60 divisions at an absolute minimum to guard the border with the USSR then again - better to focus on beating France quickly.

Your question also ties in with the 'Strategic Level' AI where countries (especially Germany) should stop trying to pick fights when it's already struggling with the ones it has. If Germany is bogged down in France it should be doing everything in it's power to delay a war with the USSR - Instead its solution seems to be to declare war with the USSR and as many other countries as it can with whatever PP it has available. This one isn't even complicated:

if (gettingAbsolutelyWrecked == true)

declareAnotherWar = false;​

Interesting observation. I played a game a few days ago and watched the front line. The AI had so many divisions re-deploying around the front line (the slow way - not strategically) that almost 75% of my troops were farting around instead of digging in or attacking.

 

[KiwiNoob]

I'm not 100% sure but I think the answer is a combination of micro and armoured schwerpunkt, neither of which are available to the AI.

Micro should be an AI's absolute trump card. A human players weakness is that they cant be in all places at all times. AI can do exactly that.

It is an option that is definitely available to the AI - it just doesn't make use of it.

 

[tom_jones]

Interesting observation. I played a game a few days ago and watched the front line. The AI had so many divisions re-deploying around the front line (the slow way - not strategically) that almost 75% of my troops were farting around instead of digging in or attacking.

The constant re-deployment definitely is a significant problem in the current version of the game from my observation. It seems that it's at least in part caused by the AI pulling troops from the area it is in progress of reinforcing, to reinforce again the area it pulled the troops from, creating some sort of self-fuelling travel loops, instead of making overall plan and sticking to it. Hopefully this is addressed at least to some extent in the next update (though, because removing it would require some deeper changes in the code, am not exactly hopeful)

 

[Secret Master]

Please don't pointlessly spam old DD threads.