[SotF] Improving the AI

Nothing explains the AI table well enough for me to go in and mess with it too much.  I think 15 is the highest.  I'm not sure why StarClad didn't make the values "X" out of 100%.  One thing that royally pisses me off is that the AI rarely researches more than what is necessary to get ships out. 

More also needs to be detailed about "WEIGHT" in both RebellionInfo and PirateRaidCompostion.  I'm not sure of quite a bit in the Gameplay.constants like PhysicsData and TargetTypeCounterData.

Good Luck.

 I really wish there was more detail to which ships, capitals, tacticalmodules, civilianmodules, research labs are built, and more control to weight how planets are upgraded.  Below, the number are all arbitrary, I just wanted numbers adding up to 100%.

For example, would be great to have buildship include all the types from FrigateAntiFighter to FrigateUtility1 and a number in percent so total is 100% for some semblance of fleet organization.

FrigateAntiFighter    8
FrigateAntiModule    9
FrigateCarrier    7
FrigateColony    3
FrigateEnvoy    5
FrigateHeavy    6
FrigateLight    15
FrigateLongRange    12
FrigateMineLayer    4
FrigateScout    5
FrigateSiege    8
FrigateUtility0    8
FrigateUtility1    8
FrigateStarBaseConstructor    2

Same with capitalships.  SOASE doesn't seem to build capitals well. They will build colony capitals, but nothing clear about what to build after.  Another "X" out of 100% built would have been nice.

CapitalShipBattleship    15
CapitalShipCarrier    35
CapitalShipUtility0    14
CapitalShipUtility1    16
CapitalShipUtility2    20

Just an example of how much control players could have in modifying the AI's choices of what to build.  Still can't seem to get them to not be total morons atm and even if researcher, don't do jack for research.

The only way to see is to experiment which is what I plan to do. I'd like an actual Hard AI that researches everything and builds a massive fleet, but we'll see.

Thanks...

Who knows how the ai table will work out, but I'm confident I can change the other stuff to suit my goals.

Also, might be useful to know, that there's a way to force the AI to research stuff and to even prioritize some research.

For this to work, you need to have spare frigate or capital ship slots (it works with modules as well, but not as effectively). Make a new ship; it's stats aren't important. Then for the research pre-reqs you put in two techs, the one you want the AI to research/prioritize, and another filler tech that simply costs too much for the AI to research. If one makes the latter into a chain, that is one which has multiple "filler" techs which down the line encompass all the actual techs in the game (ex: weapon boosters, trade boosters, etc). Over time the AI will research all the techs at the "bottom" of the chain, as it thinks that these techs will help it unlock a shiny new ship. I say over time as the AI won't be too quick. In order to avoid clutter I personally place this off the research grid, at an x-position of 19 (nets me some nice research menu overlap errors), but the AI "sees" them all anyways.

I've used this to force the AI to research things that it normally does not research, and even to research planet colonization research before even seeing a planet of that type (an example of prioritizing). In order to aid the AI keep up I've additionally made an off-grid tech that gives the AI a research speed boost and research cost reduction, to try and help it out. This doesn't allow it to keep up with a skilled human player, but it helps nonetheless.

Edit: Another thing I've done that might be of interest is I've fixed up the Intercept/Sprint ability so that ships don't go off on these weird curves but instead go in a (fairly) straight line most of the time, making that ability work far better.

So that little idea of mine worked? I didn't think about using it to give techs only the AI could research though, that's pretty clever. Might be worth experimenting with.

Thanks for this Zombies.  The playerAISharedDef looks especially enticing, I've had some success modding other games by playing with these sort of values, and I wasn't aware that this could be done for Sins, though I have to say that the options are very basic by comparison with other games.  I dont suppose you could elaborate on what each does?

Bounty- I presume this is basic pirate bounty, not applicable if pirates are off?

Tactical- Static defence.. the AI builds far too much of this too early, so it canr even build decent structures as there are too many turrets in the way.

Ships- Build a ship, fairly obviously, but does this also force the pace of supply increases?  If so then this could be a breakthrough of sorts, as the AI could be improved by a lower pace of ship building?

Research- All research, or just military research?

Civilian- This is civilian research, noncombat and diplomatic? 

Population- What can this refer to?  It ionluy occurs in the economic variant but it seems difficult to imagine that other AIs give a very limited priority to population upgrades on planets?

Starbase- Only the defensive player prioritises starbases?  Wouldn't an aggressive player want to use them to attack or to hold flanks?  Economic players to holddown fleet costs?   

Mines- It seems very odd that aggressive and research AIs want to use mines.

Planet- This might be artifact hunts and bunkers?  Or is it logistics as well?

Others- Why no others category for defensive? Others is any of the above not otherwise mentioned, so that civilian and population are missing?

The defensive AI has always seemed a poor concept to me- after all, players defend in order to do research or improve their economy, they don't defend for the sake of defending.

It might be possible to expand all this for Rebellion, as Semaz advocates.  That could be quite exciting, though what I'd really like would be the ability for players to pit their AI setups against each other by proxy, to see whose weighted random values would emerge triumphant. 

Technically, that is how SoGE's supercap system works; with *filler* -> *filler* -> *weaker supercap*. But yup, it works very, very well. While the AI might not research the AI-boosting tech right away, which I personally put to 0 cost and dev time, they tend to get it fairly quickly, or not at all. That's one thing I found absolutely fascinating; the AI does not take development times or costs into account when purchasing tech, merely whatever priorities system it's built off of. Heck, setting the techs to cost negative resources, that is one gains resources from researching, didn't change anything either.

This made me wonder, is there a research modifier that reduces the logistic slot value of a given module? As I got a very nice idea regarding that.

Edit: As well, do have a nice little "proof of concept" for the tech chaining. You see, in SoGE there are advanced fighters and bombers, which can only be built after researching the appropriate tech. The AI, being content with the standard load out, never researched this ever. As such I made a nice chain for an AntiModule frigate; *filler* -> *filler* -> *filler* -> *advanced fighters*. After around 12 minutes one of the AIs managed to research this, and following Sins randomized fighter building, had some advanced starfighters. I do have a pic or two showing this on my netbook if requested.

Depending on how you mess with this AI, you can make it as something that likes to advance at a moderate place but bunker up whatever holdings they get, so they will be a tough nut to crack. Also setting it's population spending a bit higher does wonders for this AI type.

That is awesome Lavo! This is the sort of information that the Babylon 5 mod team desperately needs. I'll post this link over there and see if we can't in turn help gather information with internal testing.

Nope, unfortunately the only thing you can do with research is globally increase logistic or tactical slots.

sorry, I'm not redoing the tech-trees for 12 races so the idiot AI will research up to tier 4 of wtfever.  StarClad put in priority levels that don't seem to do much since it is 0 or 1 instead of having the AI research through military, civilian, defense, fleet starting with the top left and working it's way down, then to Tier 1 and down the list, and so on.  or at least the priority system being a bit more than 0 and 1 so modders could set research priorities that the Ai worked toward.

/Rant Off.  just rather disappointing

@Lavo_2 Thanks for the tips. I'm not opposed to doing something like that as playing with the priority and some other ideas to trick the AI have been somewhat less than fruitful.

I plan on doing some more testing later today so hopefully I see some changes in the AI's behavior

In order to make the AI build it er... More consistently, I made it a pre-req for one of the filler ships, which works better than a chain, as mentioned previously.

You don't have to redo anything; all you are doing is adding in new research techs. See here for my hand-done reference/draft of the Empire as an example. The bottom row is a representation of the faction's "regular" techs. It does not include every single tech, ex. for a chain of techs that "end" with an ability unlocker, it just includes the ability's tech item. The filename, indicated on the top, is "RESEARCH_FORCEAI_MIL_IMP_#", with # representing the string in the boxes.

I've seen that the AI randomly researches items; their required tech level meaning very, very little.

Woot! 

I was able to get the TEC faction to research the entire Armor path with an unfair AI. It did take a while though but once it started on the path it went the distance.

From a player point of view nothing looked different too which is nice.

Well it appears I can easily influence the AI to research up to 6 unique paths.

I tried doing more by having 2 or 3 research prerequisites but the AI just didn't appear to care to research the 2nd or 3rd and only observed the 1st when trying to unlock the hidden tech. 

For example, I'm thinking of influencing the following...

TEC: Hull, Armor, AutoCannon, Laser, Crystal and Metal

Missile already gets researched because it's linked to the Superweapon.

Psi: Plasma, Laser, Beam, Shields, Hull (might link armor to hull and make the whole chain go) 

Could maybe do Antimatter or Crystal as 6th

Phase: Pinpoint Bombardment (gets pulse and beams), phase missiles, hull, armor, resources

Linking wave to the super weapon would force that research

May do culture or cargo as the sixth

I'll probably try these and see if the trend that I observed still continues then apply it to the rest of my factions.

Second round of tests with Advent versus Tec

Progression was nice with Tec players all research atleast first tier of Hull, Armor, Autocann, Laster and Missiles and atleast one Crystal and Metal research

Progression with Advent is similar with All players researching at least first tier for Plasma, Lasers, Beam, Shield, Hull and most had Crystal.

Highest researches with 5 labs built.

TEC

Hull(4 upgrades)

Armor(5 upgrades)

Autocannon(2 upgrades)

Lasers(4 upgrades)

Missile (4 upgrades)

Metal (2 upgrades)

Crystal (1 upgrade)

Advent

Plasma (4)

Laser (2)

Beam (3)

Shield (5)

Hull (4)

Crystal (2)

Pretty cool to see the AI researching this after a lot of testing where they just simply didn't research anything at all...

Question is though just how much should they research? If we try to make them research too much stuff it may negatively effect their fleet spending and such.

Also, will researcher AI advance at a much faster rate with thing method still or will all the AI quickly try to grab these in order to unlock that mystery ship they will never get?

I'm sure no one yet knows the answer to these questions, but still it seems it maybe worth while to make a modest AI improvement mod after all.

The game only allows for 2 research pre-reqs; if you put in a third it will be ignored.

This is monitored by the Gameplay.constant values; "convincing" the AI to research things won't change how much it will spend on it's research proportional to it's other spending, but will make it keep up spending, versus relegating cash to other stuff after it becomes "content" with what it has.

The research AI won't advance at a considerably faster rate than the other factions, in part due to all of them wanting the filler ships, and the aforementioned Gameplay.constant values. I've noticed that the Research AIs tend to research the "AI Booster" I put in the first however, which puts them at a clear advantage. Though with the recent change of a military boost, this may change.

Agreed. When I first delved into this in the last month or two, I was very presently surprised to see that the AI's behavior actually changed. Then to see the AI, with money being no object, eventually research every single combat boosting tech in the military tree was absolutely amazing. I mean this was unprecedented for me personally.

Lol, I did know that at one time, but just got caught up in the effort and I run debug without the pesky popups. But still even with two pre-reqs which I ended up with in that testing it still seemed to only want to follow the first one regardless.

I actually went back to Ironclads defaults and wasn't displeased with the pacing. The values are so nebulous too that I didn't want them impacting my results with some unknown/unforeseen occurrence.

Indeed, this was pretty cool. Plus the hidden frigates will never be seen making this a pretty seamless approach for a mod.

Well, we are only talking about six or so fixed research tree's. It really shouldn't impact ship building as these are all static occurrences in time and won't need to be revisited like the ship replacements. The AI also seemed to pace itself with the research while doing other things. In either case time will tell after playing some versus AI games to see how it feels.

The question is... why isn't IronClad fixing AI research behavior so they aren't worse than stupid?  if this is their way of making SP so horrible people want to deal with MP, it sucks.

I see what you mean with tagging a ship that can't be built on to the last item in a string of research.  if the AI wants to do nothing but build ships without doing weapons, armor, hull, shields, culture, metal, crystal and practically ever other research in the game and IC thinks this is good, please for the love of modding -correct their mistake.

Another way to test.  Go to a map that's pretty large or make on.  Start it going in dev.exe, locked teams, cheat happiness and walk away for 3 hours.  I find the AI after 1 hour, even a Hard Researcher, doesn't research.  Is it too HARD to research -AFTER AN HOUR!- to colonize ice planets and more than 1 level of metal and 2 of crystal?  It did ships, a little armor, some hull, a couple of weapons, but it's pathetic.

CARRY ON ZOMBIES!

Thanks Lavo_2, will test this for the B5 Mod. Is the dev.exe throwing errors when the researchsunject is positioned ways out of screenlimits?

Not at an x-position of 19; beyond that it does throw errors. If you are going to do this and want to avoid errors; make sure that the techs aren't overlapping, as that will generate errors when creating or starting a game; even though it will not have any negative effect on the AI at all.

Agreed. I also note that the AI is so bad that if a frigate/cruiser can planet bomb, it will immediately rush to bomb an enemy planet even if it isn't a siege type (as defined for the AI), and even if there are space stations which prevent loss by orbital bombardment. As one can imagine, this is a big problem for mods.

Thankfully this bit can address that. I've gotten the AI to research planet colonization techs within the first 5-10 minutes of gameplay without even "seeing" those planet types; this as you can imagine is quite useful. The AI will not research a planet colonization tech unless it first finds a planet of that type, otherwise it doesn't think that the research is necessary.

At least to some extent this does make sense. Why waste resources For any AI above Normal however the point is moot because they get resource bonuses anyways.

Good to know.

Now has anyone figured out a definite way of how to gamble priorities? I've noticed that the AI mostly prefers priority 1 over 0, but thats not a definite. Also, can we use priorities other thn 0 and 1, to better give the AI direction?

Time; it will take time to research a colonization tech, versus being able to colonize something right away. This doesn't make much of a difference in vanilla, though I believe it may in Novus Universum, which is why I force the AI to do this.

That most definitely makes sense then.

Going to test some of my CapitalShip ability re-balancing around for the AI. Should be interesting especially as they hit level 6!