Finally uninstalling this game after many years and no longer buying dlc.

  • We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
Status
Not open for further replies.
You say it's easy, and if it was as simple as 'build x building, get y' you'd be right. But you also have to train the AI to handle jobs and their priorities, move around pops, specialize planets (with their own modifiers that might make them more useful in certain ways than others), split alloy production between starbases and outposts and ships of all types... you get the idea. The system ends up too complex for the AI to handle because everything is randomly generated and they're not guaranteed anything (well, outside of difficulty modifiers but that's different).
You might have a point if there weren't mods which already did exactly this, and did it well.

Seriously they could use either StarNet or Glavius as the foundation for a competent, competitive AI.

Every issue you raise has already been handled by volunteers, working for free in their spare time. This should not be considered beyond the means of a successful company.
 
  • 6
Reactions:
You say it's easy, and if it was as simple as 'build x building, get y' you'd be right. But you also have to train the AI to handle jobs and their priorities, move around pops, specialize planets (with their own modifiers that might make them more useful in certain ways than others), split alloy production between starbases and outposts and ships of all types... you get the idea. The system ends up too complex for the AI to handle because everything is randomly generated and they're not guaranteed anything (well, outside of difficulty modifiers but that's different).

I'm not entirely sure that's accurate either.

I mean, don't get me wrong, there's a world of difference between "managing the economy" and "playing the game." Absolutely, training the AI to handle all the moving pieces in the game is an entirely different matter. For example how to spend those resources, as you say splitting alloys between starbases, outposts, etc., is a whole different matter. So is managing a fleet or doing diplomacy, etc.

But I don't mean that. I'm just talking about handling the economy itself. And there I think I would entirely disagree with you. Issues such as jobs, planetary modifiers, etc. are all turned into narrative concepts so that the game is fun for us players. But in reality these are all just numbers. They're entries on a spreadsheet. Job1 produces +2 of X resource, but costs -4 of Y resource. Modifier11 increases W resource by 5%.

Ultimately what every empire has is a production spreadsheet. You move around values within that spreadsheet and get different outcomes, and your goal is to maximize certain productive outcomes (alloys and research) while minimizing unproductive outcomes (everything else). Achieving that goal is just a matter of running the formula for any given unit of time (in Stellaris, a month). For a human that's hard, but only because the number of variables are quite large and because we think of this narratively. We see jobs and modifiers, not CellJ1 and CellM11.

The same with randomization. Randomization makes this look more difficult because, for a human, we think narratively and have certain gameplay plans. For a computer, though, randomization just jumbles up the values on the spreadsheet. It doesn't change the formula, which is the only thing that matters.
 
  • 9
Reactions:
You might have a point if there weren't mods which already did exactly this, and did it well.

Seriously they could use either StarNet or Glavius as the foundation for a competent, competitive AI.

Every issue you raise has already been handled by volunteers, working for free in their spare time. This should not be considered beyond the means of a successful company.

It shouldn't.
My point isn't that it's impossible. If the aforementioned volunteers can achieve better AI by just messing with the bare surface of AI coding (the decision weights), the AI should be better.
The problem is, it isn't profitable enough. Paradox corporate would rather sell another DLC than spend the time and resources necessary to fix the AI, because it's just difficult enough a task (untangling the deeper spaghetti code that modders don't have to touch, off the top of my head) that they don't see it as worth it. People still buy the game in record numbers and keep showing up for DLCs, after all.
 
  • 5
Reactions:
I for one will not be following you as i'm am having a lot of fun playing Stellaris. You will miss out on great mods such as New horizons. I play all space strategy games like Distant worlds universe and they all have their faults. it sounds like you want perfect but You will never find perfect and never ever will. God you fucking moaners suck all the fun out of playing computer games with your fucking moaning all the god damn time. You people do my fucking head in.
 
  • 13
  • 3
  • 1Haha
Reactions:
(untangling the deeper spaghetti code that modders don't have to touch, off the top of my head)
If two modders have already done it without touching that stuff, why would PDX devs need to touch that stuff?

Seriously, there's unlikely to be a reasonable argument about the impossibility of doing a thing which has already been done twice, and both are available for free.

Also, thinking about it a bit, the Prikki-Ti scripting seems fairly competent when I have looked into their worlds. They build Alloys and Research. They manage those resources well enough to murderize their way through other AIs.

PDX has done the correct thing themselves, it's just not used widely enough.

This should not be a difficult problem.
 
  • 8
  • 1
Reactions:
I'm not entirely sure that's accurate either.

I mean, don't get me wrong, there's a world of difference between "managing the economy" and "playing the game." Absolutely, training the AI to handle all the moving pieces in the game is an entirely different matter. For example how to spend those resources, as you say splitting alloys between starbases, outposts, etc., is a whole different matter. So is managing a fleet or doing diplomacy, etc.

But I don't mean that. I'm just talking about handling the economy itself. And there I think I would entirely disagree with you. Issues such as jobs, planetary modifiers, etc. are all turned into narrative concepts so that the game is fun for us players. But in reality these are all just numbers. They're entries on a spreadsheet. Job1 produces +2 of X resource, but costs -4 of Y resource. Modifier11 increases W resource by 5%.

Ultimately what every empire has is a production spreadsheet. You move around values within that spreadsheet and get different outcomes, and your goal is to maximize certain productive outcomes (alloys and research) while minimizing unproductive outcomes (everything else). Achieving that goal is just a matter of running the formula for any given unit of time (in Stellaris, a month). For a human that's hard, but only because the number of variables are quite large and because we think of this narratively. We see jobs and modifiers, not CellJ1 and CellM11.

The same with randomization. Randomization makes this look more difficult because, for a human, we think narratively and have certain gameplay plans. For a computer, though, randomization just jumbles up the values on the spreadsheet. It doesn't change the formula, which is the only thing that matters.
Well you see I have to disagree with that. Just because a problem is numerical doesn't mean it's easy for a computer to solve. Wanna make a software that finds the shortest way to solve a NxN Rubik's cube? It's just 6 axis to rotate and some colours to put in a matrix, a computer should solve it easily right? Except it's an NP complete problem.
I have no idea which complexity class the Stellaris economy belongs to, but it is way more complex than your average 4X/management sim. No other game does that weird balancing act between unlocking new building slots, using some of them so you can get resources to use others more efficiently, treating "you can build more rural districts" like it's a bonus in itself except it's not necessarily. And on top of that the buildings have a hard cap making the whole optimization problem even fiddler.
Not saying PDX isn't doing an awful job because they are. The way the game presents the economy just tells me nobody in game design is actually competent at playing the game. But just saying "there is a formula" does in no way mean "a computer can solve it well".
 
  • 4
  • 2
Reactions:
But just saying "there is a formula" does in no way mean "a computer can solve it well".
You're over-simplifying what @methegrate said into something incorrect, then declaring that the over-simplified thing is incorrect.

If he had said "there is a formula" and nothing more, then you'd be on solid ground, but he said something a lot more detailed than that. He talked about how Stellaris economy is comparable to balancing inputs on a spreadsheet, which is a linear algrebra optimization at worst and not an NP-complete problem -- so comparing it with an NP-complete problem would not be an accurate representation of his argument.

That said, I had no idea Rubik's Cubes could be NP-complete. Neat, I learned something today.
 
  • 5
  • 1Like
Reactions:
If two modders have already done it without touching that stuff, why would PDX devs need to touch that stuff?
Because they should solve the root issue, not the symptom.
 
  • 15
  • 1Like
  • 1
Reactions:
Because they should solve the root issue, not the symptom.
You need to show some evidence that there is a "root issue" first.

Like I said above, the Prikki-Ti already do the economy right -- this is a problem which two modders and PDX themselves have already solved.

If your "root issue" were somehow blocking competent AI, how the heck did PDX already put one in the game?

Seriously, you're not making a rational argument here.
 
  • 7Haha
  • 3
  • 1
Reactions:
You might have a point if there weren't mods which already did exactly this, and did it well.

Seriously they could use either StarNet or Glavius as the foundation for a competent, competitive AI.

Every issue you raise has already been handled by volunteers, working for free in their spare time. This should not be considered beyond the means of a successful company.
Neither StarNet of Glavius actually get the AI working, though. Both improve it, but the AI economies are still not stable in the mods.

As an example, Starnet masks the AI's failings by making it hyper aggressive when the AI still works- the early game. The AI works in the early game because there are fewer pops/jobs to manage, but more importantly with low tech levels building levels are low as well. The AI collapses once buildings begin to upgrade- it can't manage the pops/jobs once it starts upgrading buildings. By ramping up the aggression on the AI in the early game, it allows a few of AIs to snowball. This gives them a larger economic base and delays their collapse, but they still fall flat on their face once the tech tree gets filled out. You can see this more clearly with the friendship submod for Starnet- this just keeps all of Starnets tweaks but keeps the AI on default aggression settings. By 2150 it's as dead as in vanilla. Starnet actually isn't doing a great job improving the AI but is ramping up the aggression to mask the AI's failings. It just delays the collapse.
 
  • 14
  • 1
Reactions:
My preferred Game setting is Max Galaxy size - min habitability. Reason a Galaxy it has always felt small except for when you want to move your fleet across it This is because the game is all about expanding and growing as fast as possible. your species is always stable your Homefront is always secured - your threats are on the outside you gotta get big or get wiped. I wish for the game to first be less micro, second better performance, but also provide more engaging inter empire pollical management. Make so fast expansion will also lead to internal issues.
 
  • 2Like
Reactions:
You need to show some evidence that there is a "root issue" first.

Like I said above, the Prikki-Ti already do the economy right -- this is a problem which two modders and PDX themselves have already solved.

If your "root issue" were somehow blocking competent AI, how the heck did PDX already put one in the game?

Seriously, you're not making a rational argument here.
The root issue at hand here is the technical debt they have that makes this an ongoing issue.

Simply fixing the script weights won't fix that, because they'll have to then do it again when they change anything, like Glavius and Starnet do, putting us right back where we started.

Or they can make something far more flexible and adaptive that would actually solve the issue. In fact, they started to implement exactly that in 2.6, for this exact reason.

I'm not making a "rational argument" because you have no idea what is being talked about. You're looking at the end result, not the methods by which to get there.
 
  • 11
  • 1Like
Reactions:
Nice post. I've tried running several multiplayer sessions over 2020 at my local tabletop community as a replacement, and we figured out that new players manage their economies at best not better than AI. We wanted to engage in diplomacy a lot, but it's impossible if your empire falls apart in a blink of an eye. We discussed the game then, and here is some of our thoughts:
- If the game is so hard for humans, how can it be easy for AI? AI will mostly fix itself if economical system wouldn't be as punishing as it is.
- No planet can come to an equilibrium in which it will stay without excessive micromanagement. And that's because:
1. Jobs have pretty little productivity without modifiers, so without modifiers there can be little excess of any resource, and you are forced to eternally balance on the edge between deficits of different resources.
2. Production chain is more like a web, where anything you add to your empire taps into several important resources at once, and the most wicked part of that web is building slots limitation.
- Also, ships are dreadfully slow to travel through space. Exploration takes forever, wars take forever.
I scratched a mod over several evenings to fix everything I mentioned, and it turned out pretty well. You just need to make economy more abundant and casual to manage, more focused on spending your overflow, reduce pop count by 50% and crank up ship speed by 50%. Boom, that feels almost like stellaris 2 or 3.0, would play again 10/10. But that wasn't even close to being functional, and soon new dev diaries came out announcing a lot of similar changes, so before I make any significant amount of work the new patch would also do, I'm waiting in anticipation.

tl;dr: all it takes is a just a decent game designer from the crowd to make stellaris great again.

p.s. btw, stellaris common files are written in impossible-to-maintain style - there are too much repetitive code and magic constants - could be the reason the game is so hard to debug and to develop new features or rebalance or redesign old ones. Oh if only I were working at paradox...
 
  • 4Like
Reactions:
Nice post. I've tried running several multiplayer sessions over 2020 at my local tabletop community as a replacement, and we figured out that new players manage their economies at best not better than AI.
This confirms once more that 2.2 chain-of-production economy is quite hard on casual players (I strongly suspect I manage it suboptimally, too), debunking accusations that developers wanted Stellaris to be "more casual". On the other hand, for the pros, it's quite easy, if micro-heavy.
 
  • 6
Reactions:
Well you see I have to disagree with that. Just because a problem is numerical doesn't mean it's easy for a computer to solve. Wanna make a software that finds the shortest way to solve a NxN Rubik's cube? It's just 6 axis to rotate and some colours to put in a matrix, a computer should solve it easily right? Except it's an NP complete problem.
I have no idea which complexity class the Stellaris economy belongs to, but it is way more complex than your average 4X/management sim. No other game does that weird balancing act between unlocking new building slots, using some of them so you can get resources to use others more efficiently, treating "you can build more rural districts" like it's a bonus in itself except it's not necessarily. And on top of that the buildings have a hard cap making the whole optimization problem even fiddler.
Not saying PDX isn't doing an awful job because they are. The way the game presents the economy just tells me nobody in game design is actually competent at playing the game. But just saying "there is a formula" does in no way mean "a computer can solve it well".

What @HFY said. I never said that all numerical problems are easy for a computer to solve. I just meant that the Stellaris economy is a pretty basic spreadsheet problem. It's just multi-function algebra, nothing even close to the exponential complexity of a Rubik's cube.

Again, the problem in Stellaris is simply "what minimum values of W, X and Y result in the maximum value of Z?"

The part where I think we disagree is that you keep treating this as though it were an unknown problem. It isn't. This is a very well known type of problem. Everything you've mentioned does add additional layers of complexity, but only in that they add additional variables. And those new variables are linear. This isn't an exponentially expanding problem. It just adds a new line to the function series.

Honestly some of these things don't even do that. Unlocking a building slot, for example, doesn't create a new variable. It just adds one new entry onto the spreadsheet. It's a linear progression, not an exponential one or even a multiplicative.
 
  • 8
Reactions:
What @HFY said. I never said that all numerical problems are easy for a computer to solve. I just meant that the Stellaris economy is a pretty basic spreadsheet problem. It's just multi-function algebra, nothing even close to the exponential complexity of a Rubik's cube.

Again, the problem in Stellaris is simply "what minimum values of W, X and Y result in the maximum value of Z?"

Well, it's slightly more complicated than that because each production of W, X, Y, and Z require pop input, such that producing Z not only consumes the resources X & Y, but also moves pop away from producing X & Y, there's a compounding effect. Where the AI runs into trouble is when buildings start getting to level 2 and 3 and can open up half a dozen jobs in a single upgrade. Throw in that happening on 10 different planets at the same time and you end up with the AI being unable to solve the optimization problem well. Once you get short on one resource it starts the death spiral and everything crashes. Toss in conquering (or losing) a planet or two in a war and the AI is toast.

Because of this you actually don't want to solve the optimization problem for maximum efficiency, you want to have cushions- both in the form of stockpiles and also resource income/thresholds, that way the loss/addition of a couple planets or the sudden upgrading of a few buildings doesn't throw the economy out of balance.
 
  • 1
  • 1
Reactions:
Well, the game still has 15-20k daily according to steam charts, about the same as EU4, and higher than CK3 which just came out. It's still holding steady despite the terrible state the game is in currently. Certainly not too late to try and fix the game. If the AI wasn't in such a sorry state the game would likely catch HoI4 in terms of daily player base.


how many of those players play exclusively with mods
 
With popular games, even 4X ones, a relatively small number of userbase uses mods. According to Steam achievement stats, for instance, with Civ5 it's about 20%.
 
  • 2
Reactions:
Status
Not open for further replies.