v0.4.2 Release Candidate #2 Discussion

[Dilvish]

oh, and I guess I should keep in mind when trying to evaluate the AI -- when I play, I nearly always play human, as opposed to one of the stronger races -- Trith & Gysache seem to be the two strongest, & among the more popular. Certainly Humans are far from it. Perhaps I should play one of those stronger ones when assessing AI strength.

...I had submitted some C++ code changes, but I don't recall any of those being for the AI.

There was an AI C++ patch that tweaked the personality distributions.

Ah yes. I'm going to go ahead and mention I didn't get much sleep last night, and so am probably slightly fuzzy-headed, hehe.

[Sloth]

Well, I haven't gotten any feedback on the AI version that's been out for the last 16 hours or so, but I personally am fairly confident that it should be plenty challenging for any new / less-experienced players, and should overall be reasonably challenging for experienced players

I started a game using r5763 yesterday: 89, Elliptical, mature, medium, medium, High, High, High, Aggressive, Chato, 3 AI's.

As was to be expected the map was a mine field. I've only reached turn 56 yet, but the AI behavior so far was ok.

Things of notice:
- The Gysache opponent was quite strong attacking me with organic ships and conquering a planet with natives that was a little seperated from my empire.
- The Cray opponent looked quite weak having a system with no ships just next to mine, but when i tried to invade i noticed that he had reseached Planetary Barrier Shield 1 and Planetary Defense 2, which is not that easy to overcome and i had to postpone .
- I saw a single Cray troop ship flying right into a system with Natives (4 Troops), a Sentinel and a Sentry.

[Geoff the Medio]

I didn't get a response about this:

[*]Seeing a splattering of these errors:

Code: Select all

2013-02-12 00:04:29,424 ERROR Client : Variable<int>::Eval unable to follow reference: Source.Owner
2013-02-12 00:04:29,424 ERROR Client : Variable<int>::Eval unable to follow reference: Source.Owner

My guess is that this is related to the build cost of the stargate. Could you try changing it to

Code: Select all

buildcost = 500 + 50 * Count condition = Building "BLD_STARGATE"

and seeing if they go away, or are at least significantly reduced? The changed cost might be more interesting anyway...

Anyone testing a suitable game, please comment after making that change. Won't go into RC3, but would be nice to resolve.

[yandonman]

I'm afk for a bit, but I did not purchase stargates, if that matters. (i didn't get to test the suggested modification)

[Vezzra]

Well, I haven't gotten any feedback on the AI version that's been out for the last 16 hours or so

I've put out intermediate test builds, so all who don't compile FO for themselves can use them to test the recent adjustments.

[Dilvish]

I saw a single Cray troop ship flying right into a system with Natives (4 Troops), a Sentinel and a Sentry.

With the current AI I could understand that could happen once like that, but not twice at the same system. If it doesn't have visibility into what killed a fleet, it will work from inference 'strong enough to kill X' and then not try to go to that system unless it has X+margin firepower. So it may have figured its troop ship was enough to try, but shouldn't repeat the same thing at the same spot without more firepower.

It would be nice to only need that inference code when dealing with stealthed opponents (and even them, as I understand it now, should come out of stealth if they engage in combat?) I had code that tried to info from the brief glimpses of monsters that we get like that when a ship gets zapped on entry (and on our screen is now retained with scanlines), but it hadn't been working right; I think I've just figured out what was the problem with it.

Also, one of these days I'll need to get the empire Detection Range exposed to the AI for better planning as well; I'm not finding it currently.

[Bigjoe5]

It would be nice to only need that inference code when dealing with stealthed opponents (and even them, as I understand it now, should come out of stealth if they engage in combat?)

They come out of stealth in the sense that after they shoot something in combat, they become a valid target for the empire that owns the thing they shot at, but AFAIK, that empire never actually gets information on them (but maybe they do - there's a little bit of code at the end of AutoResolveCombat that seems to be compiling visibility information to give to empires that I havn't looked at closely). IMO, in situations where the stealthy ship takes part in combat, it should become partially visible to everyone with assets in the system. In tactical combat, this corresponds to a ship actually firing it's weapons, which would decrease its stealth on the tactical map, and typically give other empires in the same system partial visibility of it, which they would remember on subsequent turns. Auto-resolved combat should logically be dealt with in the same way.

[Geoff the Medio]

I had code that tried to info from the brief glimpses of monsters that we get like that when a ship gets zapped on entry (and on our screen is now retained with scanlines), but it hadn't been working right; I think I've just figured out what was the problem with it.

In what way was that info difficult to use? It should be available to AIs on subsequent turns, just as it is the player's client in order to display the scanlined fleets in such systems...

(and even them, as I understand it now, should come out of stealth if they engage in combat?)

That aspect of the combat results update code probably isn't implemented yet. Ships come out of stealth after attacking in the sense that they can be attacked back, but other empires probably won't get any additional info about them as a result of that.

Also, one of these days I'll need to get the empire Detection Range exposed to the AI...

Range, or strength?

[Dilvish]

I had code that tried to info from the brief glimpses of monsters that we get like that when a ship gets zapped on entry (and on our screen is now retained with scanlines), but it hadn't been working right; I think I've just figured out what was the problem with it.

In what way was that info difficult to use? It should be available to AIs on subsequent turns, just as it is the player's client in order to display the scanlined fleets in such systems...

It's not so much "difficulty" (I don't claim that every time there may be a bug, or failure to adapt to new dynamics, in my code it's because of any particular "difficulty"), as it is doubt whether my current way is really working right; it was written before a number of the recent visibility changes and a recent case made it look like it wasn't working right. The way I had been approaching this was in thinking that since the last time the AI glimpsed the system in this kind of case, it had one or more enemy fleets, and if I could retrieve the System object, then that system.fleetIDs list should include the fleetIDs for those glimpsed fleets (even if the fleet object retrieved by such would have 'stale' info). I had thought that was working, but then I saw some recent logs that made it look like it was not, that the AIs were getting empty fleetID lists in such a case, even though they had just lost ships in that system and so must have encountered fleets there, and I forget just why, but I thought I had reason to believe the encountered fleets were not stealthed. So, that's the "difficulty"/uncertainty. I started thinking that perhaps instead of starting with the system, I needed to iterate over all the fleetIDs the AI had info on, and check what systems they were in. I guess I'm also starting to wonder if perhaps my existing approach is working fine and I just misjudged whether there were stealthed enemies there; it's also possible that unstealthed enemies were passing through it at the moment the AI fleets landed, but I don't think that was the case. I'll have to test it out more.

(and even them, as I understand it now, should come out of stealth if they engage in combat?)

That aspect of the combat results update code probably isn't implemented yet. Ships come out of stealth after attacking in the sense that they can be attacked back, but other empires probably won't get any additional info about them as a result of that.

Also, one of these days I'll need to get the empire Detection Range exposed to the AI...

Range, or strength?

I misspoke -- Detection Strength is what I believe I would need new exposure for (I'm not finding anything for it in the Empire wrapper). Although in many cases I'd just get detection range from some particular ship, there's other times when I'd want to plan ahead about likely detectability between systems & the base detection range boost from tech is what I was thinking of when I wrote that, but I suppose I should just calculate that based on reviewing completion of specific techs.

[Geoff the Medio]

Detection Strength is what I believe I would need new exposure for (I'm not finding anything for it in the Empire wrapper).

Yeah, you'd need to expose Empire::GetMeter and ask it for METER_DETECTION_STRENGTH.

[markpi]

hi,
in the description of static multucellar hull : xenocoordination facility is required
don´t understand description of deception : if it´s not cumulative with other techs why should i take force energy camouflage ?
schwarzchild should be schwarzschild - it translates blackshield
trith feel op at start, capital is far better than other civs ( didn´t use my initial colony ship till turn 35 or so felt strange though xD)
i start at alkalurops (large) and abadonny around the corner at kilja, sslith on the other side at lockyear ( both bad ground troops )
i would like planets with bad ground troops to have at least a minor monster to take care maybe with ~10 attack,~ 20 structure
start with medium hull 3xTroops, massdriver2+3,subterranean habitation,planetary ecology + algorihtmic elegance
first ai capital scouted at turn 2 sabik- chato. i conquer with 5 mark3 about turn 25 - he still builds mark1 ( but has planetary defense network1 up, quite expensive^^)
i understand that chato was sandwiched - a george capital was at mesarthim
maybe the map generator could take care of a minimum distance between capitals ( i play 300+ systems 14 ais maniacal)
mesarthim falls ~turn 30 to the same 5 mark3 - they still build mark1
i think any capital should provide 10 rp ( -unknown ) now - hardcore mode is playing egassem actually as bad pop is quite a burden xD
maybe some ai research timings don´t work now with this doubled tec cost

black hole generator finishes at turn 50 at wedge ( delaying the other industry techs after fusion generation + industrial center )
think it is too powerful, would shift it bit to right - prerequisite theory of everything & double cost

same reward on the second ancient ruins this 1,2k pp neutronium extractor. i was so disappointed^^

tried rightclick on a planet this is perfect !

i give a screen from turn 92 - my sickest game ever - my first with this version

i´m sure it´s mentioned somewhere but pls how can i find out the ai settings i don´t see sth like home of the brave ?

this version feels great, but luck is quite a factor in the beginning that´s why i made a few proposals that might level this a bit

[AndrewW]

In the current game [5763] had the AI building Solar Orbital Generator in two different systems...

[Zireael]

Minimum distance between capitals is a very good idea.

[Dilvish]

i conquer with 5 mark3 about turn 25 - he still builds mark1

It's a little hard to figure out how that could have come about, though it must have been a combination of the AI being horribly sandwiched as you describe, and while working on that Mark I it fast-tracked a couple projects like an Oribital Incubator, and since it was hard-pressed it didn't get back to the Mark I until much later. It's a little hard to see what really could have delayed it so, but you can check the logs in such a case, each turn the AI records its total PP, and what it is building.

Bastion of the Brave is what the Cautious AIs name their capitol; if we're going to keep that up I suppose we should get around to writing a little Pedia article, though I'm not sure where it would fit in (maybe Game concepts could have a section on AI Aggression)

In the current game [5763] had the AI building Solar Orbital Generator in two different systems...

Thanks for checking, but unless those were at dim stars I expect that's probably within the range of what I have the AI do on purpose.

An AI could do that in a couple different circumstances -- if, by the time it finishes researching Solar Generators it has been unable to find a suitable Blue or White star to build at, it will build at a Yellow or Orange star. Thereafter it will continue looking for a blue or white star to build at & if it does find one it will build there, even if still in the middle of building the first (which will just be left as a backup). Alternatively, while in the middle of building one, the AI could have captured a planet on which another AI was building one. That's a reasonable thing to have backups of (some players intentionally build multiples/backups of that), so at least for the near term I'm not going to worry about writing up evaluations for the AI to trash inherited in-progress extra copies of beneficial buildings (anything that one of them would have decided to build at spot X would at least be a decent backup for another). Eventually it will be necessary so that the player can't intentionally saddle the AI with some huge unneeded project sucking up PP. but I'll expect our current playtesters to be able to refrain.

[AndrewW]

Eventually it will be necessary so that the player can't intentionally saddle the AI with some huge unneeded project sucking up PP. but I'll expect our current playtesters to be able to refrain.

Exploit loopholes in the AI? Nah, we would never do such a thing...