SMAX - The Will to Power - mod

[Induktio]

One thing I also noticed (and this may be something you've already changed) is that the AI doesn't colonize islands or uninhabited (small) landmasses. I remember Thinker AI colonizing islands as the Pirates. (Or an Ai which commits to expansion on the sea) They surrounded islands with sea bases and then built bases on it and terraformed it. It created production/economic hubs which gave the AI a lot of resources.

Simple answer is that you're not playing against Thinker AI when you're playing WTP. The mod replaces nearly everything with unit movement last time I checked, so there's nothing really similar between the AI actions of these two mods.

I noticed WTP's readme makes these kind of claims "New AI terraforming algorithm replacing Thinker's. These are few notable changes among many. [...] * Works best tiles first. * Properly selects best basic improvement option between rocky mine, farm-mine, farm-collector, forest." etc.

Unless the author elaborates on the terms like what means "properly", "works best tiles first", how these two implementations compare, or what you're even comparing them against (vanilla AI?), it's hard to justify those claims.

[Alpha Centauri Bear]

I'm not an experienced player and I want to be careful to suggest something that isn't well founded.

Come on! I don't think anyone even can measure this ephemeral "experience" in game playing! It sounds like "I am not very experienced in having fun".
😂
Please don't ever mention this again and never hesitate to share ideas popping up to your head. There is no harm in opinion expressed. The more the merrier. Free flow of information is the foundation of fan based projects and activities like this one.
You can always change your opinion later influenced by others. That is completely normal process. If you follow this forum from the beginning you'll see that every change was discussed by many people from multiple angles before got finalized and polished.

One thing I do feel more confident about is the cost of a Hybrid Forest. Which felt like it kind of broke the "flow" of building up infrastructure. If you wish, would you consider changing the cost from 36 to maybe something like 27?

I can reduce cost. I bumped them of the rationale that they are multi-purpose facilities: 1) reducing terraforming eco-damage, 2) improving forest yield, 3) +50% economy and psych. Each of these features are exceptionally useful and the last one even pays for facility maintenance. However, if you observe that nobody builds it even for that price than maybe it is too high.

Try this one: https://github.com/tnevolin/thinker-doer/releases/tag/286

Another consideration is that vanilla tends to double the cost of next similar multiplying facility in line (Fusion lab -> Quantum lab, Research hospital -> Nanohospital, etc.). The rationale is that later in the game player has more production to spend so it just reflects production power inflation cost. However, double cost may be too steep price for same exactly benefits as you noticed. So maybe they need to grow like 50-75% instead?

One thing I also noticed (and this may be something you've already changed) is that the AI doesn't colonize islands or uninhabited (small) landmasses. I remember Thinker AI colonizing islands as the Pirates. (Or an Ai which commits to expansion on the sea) They surrounded islands with sea bases and then built bases on it and terraformed it. It created production/economic hubs which gave the AI a lot of resources.

That is the one thing I try to teach them. At least in some recent versions I tested they do it. Like Pirates goes to the nearest land, etc. May need fгrther verification and and tuning, as with any AI feature.

Tech trading on an equal value level sounds good. The AI being more careful with technology also sounds good.

I feel it would be best if the AI is less willing to trade when it comes to key techs. It already seems to do this for Secret Projects, but I wonder if this can also be extended to key technologies which give things like: Reactors, weaponry that is currently state-of-the-art on Planet, air power, orbital tech, missiles, economic facilities like Tree Farms, tech that unlocks resources, etc)

It doesn't have to mean the AI will never trade them, but more that the AI values them more and takes the current geopolitical situation on Planet into account. (Like "I want two D9 techs in return for this C10 tech." "Instead of 250 energy credits, I want 2500." Or "the human player is allied with my enemy, I'm not trading these techs right now.")

An idea could also be that (AI) players which are rivals (Like being in the top 3 power chart) will make the AI even more cautious to trade key technologies because it could put their own position in jeopardy.

What should be avoided IMO is for the AI to value tech humans have as less than the same tech the other AI's have. That would create a kind of unfair feeling to put the human player at a disadvantage.

However, it would be okay for the AI to overcharge certain key techs. (If few factions have them) The argument can be made that these techs offer new options that will last throughout the game. The additional price tag could be justified as an "investment" for long term benefits.

Agreed. These are all good ideas. Customizing game behavior is very tedious and error prone thing. Nobody wants more bugs in the game. The first approach would be to try simple numeric parameters and see if they are satisfactory enough. Like plainly reduce frequency of tech trading in general and see if we get the effect. Reduce the chance of stealing tech and see if we get the effect. Etc.

Problem with key tech is - how to understand they are key tech? There is no 100% agreement in the community about it. In vanilla reactor and needlers were considered OP and so their unlocking techs. In mods this may change and so will unlocking tech value. I think vanilla AI already does good dynamic evaluation on tech value counting everything it unlocks. Which is pretty nice feature. If you noticed they sometimes offer 25-50-75-100 credits for it depending on their personal need in it. These numbers may be adjusted for inflation, of course, to match current cash flow but other than that the valuation mechanics is there.

[Alpha Centauri Bear]

One thing I also noticed (and this may be something you've already changed) is that the AI doesn't colonize islands or uninhabited (small) landmasses. I remember Thinker AI colonizing islands as the Pirates. (Or an Ai which commits to expansion on the sea) They surrounded islands with sea bases and then built bases on it and terraformed it. It created production/economic hubs which gave the AI a lot of resources.

Simple answer is that you're not playing against Thinker AI when you're playing WTP. The mod replaces nearly everything with unit movement last time I checked, so there's nothing really similar between the AI actions of these two mods.

I noticed WTP's readme makes these kind of claims "New AI terraforming algorithm replacing Thinker's. These are few notable changes among many. [...] * Works best tiles first. * Properly selects best basic improvement option between rocky mine, farm-mine, farm-collector, forest." etc.

Unless the author elaborates on the terms like what means "properly", "works best tiles first", how these two implementations compare, or what you're even comparing them against (vanilla AI?), it's hard to justify those claims.

Sorry, man. I didn't mean to diminish your AI work. After all you pioneered it!
😀

These claims were made at the time I was thinking I improved something comparing to current Thinker version at the time. It is my impression only. I may simply be blatantly wrong about my achievements. Besides, you keep modifying it so may get better in recent versions.
At some point you mentioned that you gave up doing anything AI related besides small fixes here and there. So I picked up and carried the torch. It seems you are back in this business. So I will glad to exchange ideas improving both mods, as always.

I won't go into deep details of my approach as it may take long. Here are paradigms I am trying to stick to.

Precompute all values once if possible and then feed ready results to vanilla hooks instead of recalculating them at every hook invocation. This is how I generate terraforming orders for formers, colonies and some combat vehicles. It is not 100% like that yet.

In same vein I am trying to avoid arbitrary hooks if I can do it directly. For example, I abandoned "suggest production" hook and instead just review all bases and set their production to what I need at the end of the faction turn. The effect is largely same but I don't depend anymore on when game decides to change production now, all bases are reviewed, and they all can benefit from global precomputed values, etc.

I am also trying to use standard AI priority approach to reduce coupling between different game aspects. During selecting base production phase each option is evaluated and gets its priority based on its usefulness, current needs, faction demands, etc. Then the highest priority item wins. A lot of these priority parameters are exported with "ai_production" prefixes for tweaking.

Code also tries to use dynamic values from the game whenever possible. For example evaluating terraforming improvement actually simulates this improvement, compute base, verifies if base actually likes it and, if yes, how much benefit base takes. This all using in-game functions for yield, etc. This way it becomes slightly more flexible against tweaking game rule yield variables and the like.


Update

The above alone does not guarantee better AI performance, of course. The devil is in details and exact parameter values.

[bvanevery]

Unless the author elaborates on the terms like what means "properly", "works best tiles first", how these two implementations compare, or what you're even comparing them against (vanilla AI?), it's hard to justify those claims.

I've now got some recent experience in both mods.  I don't have any opinion on this issue yet.  My Thinker games have been determined by other issues so far:

• don't leave someone alone long enough that they nuke you
• don't flank before you have the strength to do so
• don't expose a close land front as an early sea empire
• early Pirates are weak compared to Thinker AI land factions

These phenomena mostly have to do with Thinker's way of achieving "overwhelming resource production".  They're managing incredible inputs.  It's not yet clear to me how much this is due to condenser / supply crawler / borehole scumming, the inherent advantage of Transcend, somewhat nerfing the stock global warming rules, or simply Infinite City Sprawl.  If it's really "ICS wins the original game, Period, The End", well then there might not be much more to think about.

It is worth noting that these mods do not have the same basic terrain capabilities available.  Boreholes in WTP are nerfed.  I think condensers do not give a scaled bonus in their square, they give +1 nutrient.

In my own mod, the original game capabilities are allowed.  However, they're at least twice as time consuming to build IIRC, and absent the Weather Paradigm, you won't get to build them until late game.  You could do all that condenser / supplier / borehole scumming eventually if you're so inclined, but it is not going to be the one and only true obvious golden path through the game.  At some point, I may try putting Thinker's code up against my rules, to see what happens.  Not yet though.  Still evaluating Thinker in its own terms.  Well, except for I'm only playing Thinker with what might be "real" global warming damage now.  And a few other tweaks to its .ini.

[Induktio]

> I won't go into deep details of my approach as it may take long. Here are paradigms I am trying to stick to.

Okay, but generally it's important not to over-engineer the solutions. I'm not making that judgment on any specific thing here, just saying some elements in WTP code already look quite complicated. Also another consideration is that WTP presents lots of very complex parameters in its config file. It's a usability concern if it's too hard understand what those options do without knowing all the implementation details.

[Alpha Centauri Bear]

> I won't go into deep details of my approach as it may take long. Here are paradigms I am trying to stick to.

Okay, but generally it's important not to over-engineer the solutions. I'm not making that judgment on any specific thing here, just saying some elements in WTP code already look quite complicated. Also another consideration is that WTP presents lots of very complex parameters in its config file. It's a usability concern if it's too hard understand what those options do without knowing all the implementation details.

True about over-engineering. I try to avoid it if I can. Like replace it with simple solution if it comes to my mind. That works for patches. Does not for AI, though. The mere amount of things to account for is huge and AI code is bound to be thousands lines. Vanilla AI code is nightmare mess. I never ever even tried to understand it. You cannot escape this fate either. Thinker AI code is the biggest part of the project. And most obscure one too - I don't try to understand it in its entirety or integrate with it. Same for WTP, I guess. That is why I proposed to exchange ideas to see if we can simplify it.

As for obscure AI configuration parameters I never planned them to be dials users would turn the way they like and understand what is going on. This is for my own and dedicated testers only use. I sure can use separate file or keep them in the code but why bother? They are clearly marked as AI parameters at the end of the file. They are mostly described as well. I don't see an issue.

[bvanevery]

You cannot escape this fate either. Thinker AI code is the biggest part of the project. And most obscure one too - I don't try to understand it in its entirety or integrate with it. Same for WTP, I guess.

So far in the WTP sources I've tried to partially understand one function.  The one I think is likely responsible for the "keep spreading forever" policy.  Which I know is strategically wrong.  And of course, being only interested in partial understanding, I haven't even remotely proposed an actual working solution for anything going on in there.  Only the proposition, that effort should be focused on this one function.  That it's a locus of problems and probably represents some kind of low hanging fruit.

As for obscure AI configuration parameters I never planned them to be dials users would turn the way they like and understand what is going on. This is for my own and dedicated testers only use.

Valid design goal.  It makes it possible for someone as hardcore as myself, to do playtesting, consider very carefully the biggest problems, and look for a knob that might solve the problem.  Maybe even feed in a value and see if it actually does something better.  Without such knobs, and a list of "choke points" in the code where it would indeed be profitable to look for answers, I surely wouldn't bother.  I mean, I've got a DX12 3D engine project to draw my attention.  I don't need to be "deeply in the weeds" for something that will never make me a dime, especially considering my ongoing state of real world poverty.

So yeah.  Knob it up for the rare experts that come along.  It makes it possible for someone else to follow your work.  Yitzi, in particular, never did that.  Probably never conceived of a need to do that.  Got busy with his own life.  And hey guess what, other people trying to pore over obscure things you did in machine code, is not sustainable and not an invitation to follow on the work.

Even my work on the original game, often didn't proceed with understanding.  It proceeded with empirical observation and a helluva lot of playtesting.  I've "hacked" my way to many solutions in the original binary.  Better map generation, would be an obvious self-contained example of that.  It preceded me actually starting SMACX AI Growth mod.

Another example would be the "Deirdre wants Locusts before going Green" bug in the original binary.  I hacked my whole SE table around that problem.  Makes my SE table look stupid, unless someone bothers to actually read my design notes.  Or I verbally explain it to someone who's been in disbelief the umpteenth time.  :-)  I kinda miss vonbach that way.

I hacked my way out of the stock AI being stupid about SUPPORT...

It's kinda like the game's AI is a "sort of" ersatz neural network, with weights.  I have manually provided a pile of weights.  Other game devs might just say pff, data driven design.  As in game dev 101.

Or from the 3D graphics / asm optimization jock standpoint, you optimize a real system, for its real behavior.  Theory on paper doesn't work.  It may inform your heuristics as you try to optimize stuff, but real world results are king.  You have some commercial benchmark, you either win it or you lose it.

[Alpha Centauri Bear]

Valid design goal.  It makes it possible for someone as hardcore as myself, to do playtesting, consider very carefully the biggest problems, and look for a knob that might solve the problem.  Maybe even feed in a value and see if it actually does something better.  Without such knobs, and a list of "choke points" in the code where it would indeed be profitable to look for answers, I surely wouldn't bother.

Err... It is not a secret. You don't need to guess. Just ask me and I'll be glad to share.
😀

I would appreciate any attention to this. I just meant I didn't expect anyone would be paying attention. Therefore, they are not explicitly documented.

[bvanevery]

Hmm.  Perhaps I should elucidate my long held and battle tested Rules Of Engagement about open source.

Asking devs about stuff, is typically pretty slow.  It also has political risk.  You can piss off the dev.  It doesn't actually take much with typical open source devs working on things.  Although you and I have a track record, there's still the risk of pushing you too hard on something.  You can take a lot, but in the past I've observed, even you have limits.

Thus, in the vast majority of cases, I don't ask.  I find.  I search.  I RTFM and study the code.  I also assess how much energy I'm personally willing to put into something.  I'm not going to get into a big dev conversation with somebody, leading them to the expectation that I'm going to do a lot of work, until I've actually decided already that I'm going to do some work.  In open source, it's better to "do your own recon", look over the horizon, and quietly disappear if it looks like it's gonna be a hellhole.

There's also just the reality of, how much do I have going on in my own life?  I've given energy when I've had the free time and attention span to do so.  I'm very aware of how that can evaporate.  Therefore, I don't start creating expectations that I can't finish.

Example: I say I barely looked at the code for 1 function, because I don't want to give any impression at all, that I spent any more energy than that, or am committed to doing so.  With a development attitude like that, there's not that much for me to be asking about!   

I mean we all wish someone else would write the AI, or provide the 3D engine, or cook up the art assets, or get the legal papers drawn up.  The reality is, I either find ways to get that stuff done for my own, commercially viable work... or I get to keep living out of a car, until I'm literally old and literally die.

So although day to day, part of me might want to get more engaged to certain things, the other part of me is doing "triage" of how I can afford to spend my time.  Digging further through the gory details of that 1 function, ain't it right now.

What I actually ended up doing was a Thinker assessment.  Plenty time consuming, but not that much cognitive load to it.  I mean I need to play some kind of game anyways, and I can do AARs in my sleep.

It does make me wonder about the agenda of "more efficient" AAR production, and what the commercial import of such might be.  I haven't put brain cells into it and don't have any serious answer yet.

It's up there with the "how do you run a web forum democratically" problem.  Like I recently got enlightened about some of Paradox's past modder debacles.  There was something called the Magna Mundi mod for Europa Universalis III.  The rest of the story about that is epic.  Not "Diaspora suicidal" epic, but involuntarily committed epic.

Too bad I'm not rich.  As a younger man, a friend of mine and I joked we were supposed to be fops by now.

[dino]

> "Keep spreading forever" policy

AI maybe should dedicate resources to rush infrastructure in new bases and always send a former, or two with midgame colony pods so they can get off the ground as fast as possile and go for midsize production centers instead of fully fledged bases.

But you haven't really made a good case why it's a bad policy in general. Yes the game may end before these bases will took off, but what if the game manages to last 300 turns ?
There is not a single unlimited production piece, that provides better return on investment than a new base, so you are asking for AI to stop playing competitively at some point, so it's not too annoying to finish it off.

As for annoyance aspect, it's easily solvable by not playing on huge maps and didn't Inductio already implemented an adjustable AI expansion limit because of your complaints ?

[Alpha Centauri Bear]

Asking devs about stuff, is typically pretty slow.  It also has political risk.  You can piss off the dev.  It doesn't actually take much with typical open source devs working on things.  Although you and I have a track record, there's still the risk of pushing you too hard on something.  You can take a lot, but in the past I've observed, even you have limits.

You are thinking about it too much. And you also somehow mistake developers with mean people. Not all of them are mean.
😉

I wouldn't worry about it. This is not any large size project and I am not under any kind of pressure. So will be glad to talk about anything - as I said and I meant it.
Not expecting any commitment from anybody either. Not even from myself.

[bvanevery]

Not expecting any commitment from anybody either. Not even from myself.

Well, even that is not something you just find out from people.  "So, are you committed to what you're doing?"  Gets on people's nerves fast.  I don't ask about that.  I look at how many commits they're making to their code repository, whether they respond to questions put to them, and critically - if I do work, do they do work?

You're fine.  I'm just saying, I don't ask about this sort of thing.  Which leaves people's level of commitment, somewhat in the dark.

Try to understand I've lost literally man years of my life to $0 open source developer politics.  I'm very gunshy about doing lots of work where I get burned again.  Maybe almost to the point of PTSD.  "Overthinking it" as you claim.  Working on SMACX AI Growth mod all by myself, is the logical conclusion of such a history.

[bvanevery]

But you haven't really made a good case why it's a bad policy in general.

I don't need to.  It's clearly broken policy in WTP right now.  I've written tons of AAR-like posts on the subject.  I've got issues filed in the Issue Tracker.  Fix what I've QA'd.  Then worry about whether it's a good/bad policy "in general".

so you are asking for AI to stop playing competitively at some point,

It's not playing competitively now.  Thinker is competitive now.  WTP isn't.

didn't Induktio already implemented an adjustable AI expansion limit

I haven't tried turning that knob.  One reason not to do it, is Thinker seems to derive its strength overwhelmingly from condensers, supply crawlers, and thermal boreholes.  One of my test games, there were like 80 crawlers in play.  I thought, "If only I could find a way to kill those things?"  That's area based growth, not vertical growth.  I'm slightly concerned that Thinker might not actually know how to grow big cities, absent these exploits.  "Concerned" meaning, my next testing procedure is throwing what I consider "real rules" on top of it.  Instead of that clearly golden path exploitable original game stuff.  I really don't like "the boreholes only" game.  Very much offends the Deirdre tree hugger "forest and forget" in me.

WTP is similar to my modding work in that it makes "real rules" for just about everything.  But most especially, about terrain.

because of your complaints ?

Hardly.  Could have been someone else's identical complaints for all I know.  Could have been Induktio deciding it's The Right Thing To Do [TM].  But we generally speaking, have not previously given any real feedback on each other's work.  We got off on the wrong foot at the beginning, a few years ago.  Fundamental disagreements about AI faction SE choices, global flooding from AI boreholes, and utilization of [power tech wealth growth] weights in alphax.txt.  Those first 2 issues, well nowadays Thinker has knobs for them.  I'm still checking on how those knobs work, to see if they do what I hope they do.

I don't know if the 3rd issue is even an issue anymore.  If it is an issue, it'll affect how dumping my mod on top of Thinker goes.  If factions research stuff pretty much like I expect, based on long experience with my own work, then there's no issue.  If factions mysteriously cheat and beeline for certain techs, then there's an issue.  Those weights are supposed to be a racetrack for how blind research works, for how easy or difficult anything is, depending on your choices.  It's not supposed to be some advisory nice idea that you can just blow off, and instead implement your own internal directed research algorithm.  You make a choice about general area you want to research in, and then you're stuck with the probabilistic consequences of your choice.  That's the game.  And I've spent years fine tuning that particular game.  Very exact and disciplined regimen of what Explore Discover Build Conquer mean, compared to the slop of the original game.  I want the AI to play the same game I'm playing, and not have some special cheat put into the binary.

[dino]

I agree that techs ai weights are much better handled in the alphax.txt, than in the gamecode, but making his tech weights optional is like the first thing he made due to your feedback at the very beginning. Just like with techs weight, you were very vocal about being annoyed about infinite expansion at the time he added this option. Even if you don't get along it doesn't mean he would discard an idea only just because it came from you.

[bvanevery]

I agree that techs ai weights are much better handled in the alphax.txt, than in the gamecode, but making his tech weights optional is like the first thing he made due to your feedback at the very beginning.

Well if that happened, I never noticed, because I stopped following Thinker's development.  I don't see anything about tech weights in Thinker 3.1's .ini.  Alternate tech cost to research, yes, but not anything about [power tech wealth growth] for blind research.  I suppose I could play catchup right this minute and read the entirety of Changelog.md.

Ok, just did that.  I don't see anything obviously relevant.  Moving along to Details.md.

Ok, read that too.  I don't see anything.

I'm not planning to dig through Thinker's source code at this time.  What actually matters is the test game I'm about to run, and whether blind research is "noticeably off" somehow.  No point worrying about it until I actually see something.  And I'm aware that even the original game binary, did dirty tricks to get the AIs to the Hunter-Seeker Algorithm.  There are a couple of other dissassembled goosings that some people have discovered.