HOI4 Dev Diary - Patch 1.3.3 Update #2

[Real Strategist]

"Hi everyone, I'm going to be talking about all the optimization work I did for the patch. We are of course planning to improve it further more in the next patches, however in 1.3.3 the overall speed up is about 14-18%."
...
As I previously said, we are still working on late-game performance. I've spent only 1,5 week on it so far, and we already get quite much of it so you can expect us to keep doing improvements like this in future patches."

You did all this optimising in just 1.5 weeks? And you get around 15% more performance out of HoI 4?
Holy smokes!
This is amazing!

They should double your salary and give you an other month or two for optimising.

If you find some time to remove ghost ships and the tedious 'coward bug' from game I promise to name my secound child

TOMAS +THE-GREAT +(my last name)

The AI is still not attacking undefended territory on large borders atm. I did a game as canada a couple days ago. US declared war on me and did nothing. Even though 90% of my border was completly undefended. I took Washington and got an archivement. 25 vs 165 divisions.

The AI seems to be unwilling to attack undefended or weak defended territory when in overwhelming force.

It seems to constantly rotating from north to south or west to east instead which leaveshuge gabs and troops with 0 org on frontline.

 

[Real Strategist]

'Perhaps it would now be worth considering lowering the distance that was changed in 1.1 from 8 provinces down to 6 since that is a minor factor compared to what you changed later.' (from Dalvin ps: quotes doesn't work atm)

Hell no!

I would remove strategic redeployement completly out of the AIs 'hands'. There is allready a button called 'allow strategic redeployement'. So why is in game when Ai does it anyway?

It drives my crazy when AI doing it on large fronts. I played as PRC today and fought China+Japan simultaneously. I had just one good general (Mao) so I got a single but very long frontline barely covered with troops. I takes at least one month to get from Mengukko to Xibet Xan Ma but the AI thought it would be a good idea to move soldiers from hotspots (where I finally fought a gap into enemy lines) into completly abandoned frontiers.

We don't need more SRDs we need controll of them. Because I can allways SRD my troops manually but I can't prevent AI from SRD my troops. The only things that work are shorten the fronline or completly delete it and all plans attached to it.

 

[dsturnbull]

You did all this optimising in just 1.5 weeks? And you get around 15% more performance out of HoI 4?

Let's just hope he didn't exhaust all the low hanging fruit yet

 

[adam_grif]

You did all this optimising in just 1.5 weeks? And you get around 15% more performance out of HoI 4?
Holy smokes!
This is amazing!

They had begun work on some of this stuff before last year ended. And even though this patch only just came out (in beta), they have already begun working on the next patch to come later. Sometimes they work on things that aren't quite ready to ship when a patch comes out, so it gets shifted to the next patch instead.

 

[dsturnbull]

They had begun work on some of this stuff before last year ended. And even though this patch only just came out (in beta), they have already begun working on the next patch to come later. Sometimes they work on things that aren't quite ready to ship when a patch comes out, so it gets shifted to the next patch instead.

Don't mansplain this. The dev stated a week and half in the OP.

 

[Dalwin]

'Perhaps it would now be worth considering lowering the distance that was changed in 1.1 from 8 provinces down to 6 since that is a minor factor compared to what you changed later.' (from Dalvin ps: quotes doesn't work atm)

Hell no!

I would remove strategic redeployement completly out of the AIs 'hands'. There is allready a button called 'allow strategic redeployement'. So why is in game when Ai does it anyway?

It drives my crazy when AI doing it on large fronts. I played as PRC today and fought China+Japan simultaneously. I had just one good general (Mao) so I got a single but very long frontline barely covered with troops. I takes at least one month to get from Mengukko to Xibet Xan Ma but the AI thought it would be a good idea to move soldiers from hotspots (where I finally fought a gap into enemy lines) into completly abandoned frontiers.

We don't need more SRDs we need controll of them. Because I can allways SRD my troops manually but I can't prevent AI from SRD my troops. The only things that work are shorten the fronline or completly delete it and all plans attached to it.

Such a wide front means that it could easily decide to have some unit walk 12 or 15 provinces. He will arrive weeks later and the front will have shifted so he will begin walking again, perhaps even back to near where he started. The SR is not the problem. The number of times it thinks it has a hole to plug and the amount of time units spend not at the front because they are supposedly trying to reach some hole, these are the problems. The longer it takes to get there the longer that unit is not in the line. This can easily snowball to where very few units are on the line and those that are don't think they are strong enough to attack.

They need to stop the walking of long distances.
They need to prioritize the shuffle so that units move to spots that are nearer.
They need to recheck units in transit periodically to see if their destination is still a valid one.

These things will fix the problem. Wide fronts and never using trains will only make the problem worse instead.

 

[grandad1982]

Such a wide front means that it could easily decide to have some unit walk 12 or 15 provinces. He will arrive weeks later and the front will have shifted so he will begin walking again, perhaps even back to near where he started. The SR is not the problem. The number of times it thinks it has a hole to plug and the amount of time units spend not at the front because they are supposedly trying to reach some hole, these are the problems. The longer it takes to get there the longer that unit is not in the line. This can easily snowball to where very few units are on the line and those that are don't think they are strong enough to attack.

They need to stop the walking of long distances.
They need to prioritize the shuffle so that units move to spots that are nearer.
They need to recheck units in transit periodically to see if their destination is still a valid one.

These things will fix the problem. Wide fronts and never using trains will only make the problem worse instead.

Even on small fronts it can be hard watching where divisions are heading. Seeing one head to a spot a few provinces away to get there after the front has moved forward, to then deside to head off in a similar fashion again is pretty frustrating. Once this kind of thing happens if seems to magnify itself quite quickly. The other thing that is annoying is when multiple attacking divisions all move to the same province rather than some of them switching to attacking/supporting other adjacent provinces.

I often find myself just selecting an entire army and hitting hold periodically so they all reroute to (usually) nearer destinations. It's minimal micro but for a very quick action can be pretty effective if your attention is split.

 

[Dalwin]

Even on small fronts it can be hard watching where divisions are heading. Seeing one head to a spot a few provinces away to get there after the front has moved forward, to then deside to head off in a similar fashion again is pretty frustrating. Once this kind of thing happens if seems to magnify itself quite quickly. The other thing that is annoying is when multiple attacking divisions all move to the same province rather than some of them switching to attacking/supporting other adjacent provinces.

I often find myself just selecting an entire army and hitting hold periodically so they all reroute to (usually) nearer destinations. It's minimal micro but for a very quick action can be pretty effective if your attention is split.

There are multiple ways to "reset" an army manually so they pick new destinations. As you say they usually pick ones that are closer than what they had been doing. I do ti by selecting the entire army (click on general) then control + click on battle plan line on map (as if you were adding them to it, even though they are already part or it. One doesn't need to do this for the entire army. You can also select just some units and do this so they pick new orders. Doing it for only some allows others to continue their attacks.

 

[Aodhan_]

Uruguay is just used for -handsoff games, since they are unlikely to get involved in anything.

Maybe it would be more beneficial to use a country that gets involved so you can actually test air, naval and all land combat features such as battle planer and them shuffling ( Fu***ing not going to say it )divisions. Not trying to be ignorant but to me your testing seems to be hands of testing or playing silly countries.

 

[Archangel85]

Maybe it would be more beneficial to use a country that gets involved so you can actually test air, naval and all land combat features such as battle planer and them shuffling ( Fu***ing not going to say it )divisions. Not trying to be ignorant but to me your testing seems to be hands of testing or playing silly countries.

Handsoff testing is handsoff testing. That's why we have a -handsoff command that automatically puts you in uruguay, so we can see how the rest of the world progresses. If we want to see how the AI does things, we observe it directly. -handsoff is mostly used to run overnight games and to gather statistics (i.e. "Japan beats China before 1942 8 times out of 10", "Germany takes out France before 1941 99 times out of 100" etc.)

 

[Aodhan_]

Handsoff testing is handsoff testing. That's why we have a -handsoff command that automatically puts you in uruguay, so we can see how the rest of the world progresses. If we want to see how the AI does things, we observe it directly. -handsoff is mostly used to run overnight games and to gather statistics (i.e. "Japan beats China before 1942 8 times out of 10", "Germany takes out France before 1941 99 times out of 100" etc.)

My apologies Archangel my post doe's seem abrupt no excuse for ignorance. I do know why Paradox doe's hands off testing for, but seems to be more hands off than hands on from my perspective and in doing so letting stupid things through the net. Then you have to rely on player's sending saved games with said bugs/glitches or what ever, maybe this saves Paradox time somewhere but increases frustration with players. It would also be useful to experience said problems first hand, and not through multiplayer.

 

[Archangel85]

My apologies Archangel my post doe's seem abrupt no excuse for ignorance. I do know why Paradox doe's hands off testing for, but seems to be more hands off than hands on from my perspective and in doing so letting stupid things through the net. Then you have to rely on player's sending saved games with said bugs/glitches or what ever, maybe this saves Paradox time somewhere but increases frustration with players. It would also be useful to experience said problems first hand, and not through multiplayer.

We have two embedded QAs whos entire job is to do exactly that, we have semi-regular playtesting days where the entire team plays through specific scenario setups ("As Germany, go down historical path", "As Germany, try to capture as much of Europe as you can before getting into a war with another major" etc.) in singleplayer.

 

[hillcf]

We have two embedded QAs whos entire job is to do exactly that, we have semi-regular playtesting days where the entire team plays through specific scenario setups ("As Germany, go down historical path", "As Germany, try to capture as much of Europe as you can before getting into a war with another major" etc.) in singleplayer.

I do have a concern about testing vs QA.

QA can be very good at finding out if the game plays as planned. So in historical mode, make sure that Germany does not attack Switzerland. But that is not testing, although it does have the ability for the QA to see some of the issues with data content within the game.

Testing is much more concerned about ensuring that the code in the system is changing or modifying the correct information, or calling the correct routine with the correct parameters. Somehow, I think that some of the issues we are seeing are because of detail within the code not working correctly.

As I was told when I started programming 47 years ago --
1. Test, Test and Check.
2. Always get someone to check your code, because you will never spot the silly errors.

I did it then, and then ran a technical support team for many years. Even as an IT Manager I still had to do it on users, developers, suppliers and on anyone else who had issues.

 

[Aodhan_]

We have two embedded QAs whos entire job is to do exactly that, we have semi-regular playtesting days where the entire team plays through specific scenario setups ("As Germany, go down historical path", "As Germany, try to capture as much of Europe as you can before getting into a war with another major" etc.) in singleplayer.

I do hope Archangel that the two embedded QA's don't play at speed 5. It is also good to know that you play single player for play testing.

 

[Dalwin]

I do hope Archangel that the two embedded QA's don't play at speed 5. It is also good to know that you play single player for play testing.

Isn't one of those QA's Daniel? Unless his play style is different on WWW than the rest of the time, he seems to greatly prefer speed 5 and controlling things with broad strokes thereby missing a lot of detail.

 

[kelestra]

disappearing frontlines: I was not aware that this bug was back again. If you are able to reproduce it; please report it in our bug forums and tag a dev in the report.

merging front lines: I fear you will need to be a bit more specific here; which part of fronts merging are you having problems with? If it is only how AI handles it on its end (not for the player), then I am aware there are some issues.

division shuffling: This is going to be a farily high priority in the coming months or so.

embarking into aread with 0 sea dominance: We are aware that this happens, but a good and proper solution is pretty complicated to accomplish for this. I will see what I can do.

First of all: Thank you for your answer! Much appreciated!

Then: the answers!

1. Merging frontlines: Imagine me playing GER and executing Operation Barbarossa. I draw three frontlines in Poland. A northern one towards Leningrad-> Archangelarsk, in the middle one towards Moscow->Ural and in the South one towards Stalingrad/Baku->Astana. Let it just be 2 or three steps till the final destination. If you start the attack and do a "hands-off", it happens quite often that after a few provinces into SOV territory the northern frontline reaches till the Black Sea, the southern one to the Baltics for example. therefore many units get thinned out and scattered all over the GER-SOV-frontline, and more important, the attackangles (not only the visualisátion of the arrow) and order of provinces to progress are changed. Most likely to your disadvantage, because you don't want your northern army to push towards Woronesch, but follow more the Baltic coast towards Leningrad.
2. Disappearing frontlines: Same scenario as 1.. Imagine the detail of the southern frontline towards Stalingrad and beyond. Of course you don't want to ignore the Krim and some SOV divisions to cut of your frontline from there. So you take one part of your southern force to secure Sevastapol and perhaps afterwards towards Baku or so. So what happens everytime is, that when the force passes Mykolajiw, the attack arrow towards the krim just disappears. ALl units assigne to it will rather stop or join the Stalingrad attackangle. Very often it happens, that when I draw attack angles deep into SOV territory, the whole battleplan disappears beyond Moscow e.g..
3. Embarking into enemy controlled sea areas: I thought you could just add a button to battleplans/armies to force a land route in their general movements (perhaps by ingoring sea provinces as passable terrain?!). Or introduce a check on AI pathfinding that just uses the current sea dominance to add weight/movementcosts for this path, just that the algorithm is not calculating sea-provinces as the route with the optimal movement costs.Sometimes I got so desperate to LL all my convoys to my Axis so that there are none for embarking...
4. That you are aware of divisions shuffling and working on it is comforting. It's a real struggle to see your battleplan looses all it's effectiveness because the starts shuffling 10 units from north to south and ten others from south to north, and then again and again. Instead of shifting units in a cascade kind of style or not at all...

 

[orwell]

I really enjoy reading about the details of how things work within the game. Thank you @SteelVolt, @TomaszKowalczyk, and @podcat for sharing.

 

[Gnollu]

Where is my new dev diary?
I want it, I need it I cannot bear without it
@podact and rest of PDS... You are the only reason I want to be alive every wednesday

 

[SilentNinj]

Don't think I didn't notice that cheeky Farscape reference

 

[joreri508]

You are talking about having the formulas determine what from history books? The models you use in examples are for watching statistical data change based on specific "text" searches from historical documents. How could these mechanics be used by the AI to improve its choices in a game? The dynamics the AI would be responding to would not correlate to actual historic activities or events.

Machine learning, for a self-developing AI. The same way we humans and other animals learn from experience. There are powerful algorithms for this now. The problem with a self-learning AI though, would be to set boundaries for realism, as you say.