AI Development Discussion and Feedback

redrum

@Alexei-Svitkine Some good examples here. Quick thoughts:

The AI struggles with 3 move planes and maps without carriers and need ssome additional checks around those.
The AI doesn't quite understand the neutral relationships between the Central Powers and Communists.
The AI only has limited support for 0 move units (trenches).
Guessing the unloaded units was some sort of movement bug

Alexei Svitkine

I noticed some very buggy AI behavior. On WAW, Hard AI playing the Japanese factions was sending a bomber to attack a well defended ANZAC territory. The bomber just died as expected. And WAW uses low luck, so it was a guaranteed suicide. This happened twice with a bomber over different turns another time with 2 fighters.

(Getting a forum error adding a screenshot. Can try later.)

This is with latest prerelease. Save:
waw_buggy.tsvg.zip

Frostion

I also get error when trying to add screenshot. Tried to do it like 15 hours ago. What's up with that?

bluelionman

Big World 1942 v3

I'm not sure if this is the right place to post this but wanted to offer some feedback. I play the AI a lot and find I beat it even on Hard but when I play a human I usually lose (with the exception of my sibling whom I usually beat).

I played a really long game against the AI where I was only 1 country - all the rest on Hard AI and I tried really crazy things at the start as I was testing the theory you can do anything and win against the AI. Well there is no way I should of won - I was losing big time the Axis (AI) had taken Russia and the whole continent was there's between Germany, Italy and Japan... also Italy had all of Africa. I had loads of powers for UK and just kept buying infantry and bombers stopping the Axis having a navy to get to UK or USA and having enough defend against parachute drops of dudes.

This stale mate went on for ages then in round 40 something I noticed Germany had moved everything out of its capital I'd no idea why but I pounced had 6 bombers drop 6 British infantry. His AA shot 2 down but 4 landed and I took his money over 100 PU's.

Now any human would not make that mistake again - he had 55+ units off all types 1 territory away but every German go he only put half a dozen units in to take it back and moved no more in on non combat to re-in-force. This meant I kept retaking Germany every round for the next 4+ rounds and its money so it never got to build and Britain got to spend near on 140 PUS every round - so of course I came back and won.

So somehow the AI does not realise the significance of its capitals and how losing them means it loses its PU's. Can anything be done to address that?

Also I found the USA as an AI is dead useless it just seemed to buy transports and infantry and sail up to Alaska and pretty much drop them off and go back and that's all it really did most game and drop some off up and down that coast when no chance of holding off Japanese units and leaving its navy to get sunk by planes over and over again. The US could of helped so much more having the most PU's income of the allies early on but seemed to do nothing much - no plan whatsoever.

I don't know how AI is coded and it must be real hard but is there anything being worked on to help it do better?

I don't wish to moan I love the game just wanted to feed back my experiences in the hope it may help those behind this great game online. Thanks all.

pithief

Of note, I believe the new 2.1.20365 engine does not build new factories anymore, at least under certain maps. Previously played through a couple games under the scenarios The Rising Sun (Map v. 1.9.3) and 270BC (map v. 1.7.0) using the 1.9 engine and the AI would build plenty of factories or cities. We all updated and now they don't.

Hope this is the right place to put this. Also, ya'll have done great things with and for this game!

TorpedoA

Hi there
I have a unit called TrenchInf.
Supposed to be like a Infantry which is entrenched and in a extreme defensive position so they cannot attack immediately so to speak, so i want them to be a noncombatmovement infantry.
At first it looked like this:

<attachment name="unitAttachment" attachTo="TrenchInf" javaClass="games.strategy.triplea.attachments.UnitAttachment" type="unitType">
<option name="movement" value="1"/>
<option name="attack" value="0"/>
<option name="defense" value="3"/>
<option name="isLandTransportable" value="true"/>
<option name="transportCost" value="4"/>
<option name="canNotMoveDuringCombatMove" value="true"/>
<option name="isConstruction" value="true"/>
<option name="constructionType" value="FieldFort"/>
<option name="maxConstructionsPerTypePerTerr" value="20"/>
<option name="constructionsPerTerrPerTypePerTurn" value="2"/>
</attachment>

Then i noticed that the AI seems to never non-combat-move it by sea transport.
Then i thought that it could be something about infrastructure, so i now changed it to something like a normal infantry unit:

<attachment name="unitAttachment" attachTo="TrenchInf" javaClass="games.strategy.triplea.attachments.UnitAttachment" type="unitType">
<option name="attack" value="0"/>
<option name="defense" value="3"/>
<option name="movement" value="1"/>
<option name="transportCost" value="4"/>
<option name="isLandTransportable" value="true"/>
<option name="canNotMoveDuringCombatMove" value="true"/>
</attachment>

Unfortunately the AI still doesnt move this unit by sea transport in noncombatmovement round.

This is all Fast AI.

Any ideas?

ff03k64

I am wondering if it is possible to have the AI from previous version. I.E. If i design a single player map balanced for the current AI, I probably want a way for it to still be relevant in the future.

TorpedoA

About "Show Hard AI Logs"

Where do those Efficiency numbers come from?
For example Infantry
Movement=1
Attack=1
Defense=1
PU-cost=4
HP-Efficiency=0.4
Att-Efficiency=1.5
Def-Effiecieny=1.75

How?

Mora

@pithief said in AI Development Discussion and Feedback:

Of note, I believe the new 2.1.20365 engine does not build new factories anymore, at least under certain maps. Previously played through a couple games under the scenarios The Rising Sun (Map v. 1.9.3) and 270BC (map v. 1.7.0) using the 1.9 engine and the AI would build plenty of factories or cities. We all updated and now they don't.

Hope this is the right place to put this. Also, ya'll have done great things with and for this game!

I second that. Unfortunately, AI doesn't build factories any more like it did. I am a WAW player. It's no fun anymore...

Frostion

@Mora said in AI Development Discussion and Feedback:

@pithief said in AI Development Discussion and Feedback:

Of note, I believe the new 2.1.20365 engine does not build new factories anymore, at least under certain maps. Previously played through a couple games under the scenarios The Rising Sun (Map v. 1.9.3) and 270BC (map v. 1.7.0) using the 1.9 engine and the AI would build plenty of factories or cities. We all updated and now they don't.

Hope this is the right place to put this. Also, ya'll have done great things with and for this game!

I second that. Unfortunately, AI doesn't build factories any more like it did. I am a WAW player. It's no fun anymore...

Is there any dev willing to look into this issue? Or at least just identify the issue?

I have no idea if this issue is posted at Github.

By the way, a bug like this really hinders map development, besides messing with the single player experience, especially for maps that utilize the AI in some way.

Mora

https://github.com/triplea-game/triplea/issues/7353

ff03k64

I am wondering what the priority is for getting the AI to handle resources besides PUs very well. Thanks!

LaFayette

AFAIK the biggest priority is for factory purchases to be brought back.
@redrum , do you remember the background on why AI stopped purchasing factories?

ff03k64

@LaFayette I was more wondering like short term, mid term, long term.

Because I know a lot of maps don't use them, some like Warcraft the AI can handle them ok based on how they are implemented. Others like fuel on iron war they just smash their heads into the wall.

LaFayette

@ff03k64 I don't want to discourage discussion here, however TBH our labor capacity is critically overstretched these days. Therefore certainly not short-term, probably not even medium term

ff03k64

@LaFayette That's fine. I am just trying to decide how to design my map, and if this is a ways off, i will probably use triggers to get desired performance out of the AI.

I don't want to duplicate requests, do we know if it is on the feature request page? I am terrible at search, and don't want to go through the hundred posts that i get to see if they are remotely what this is.

LaFayette

I think AI handling alternative resource types exists as a feature request, but TBH I'm not completely sure it is.

beelee

@LaFayette said in AI Development Discussion and Feedback:

I think AI handling alternative resource types exists as a feature request, but TBH I'm not completely sure it is.

I don't see it listed. @ff03k64 you can make a request one thread above this one.

Ahh I see you already made one. Please disregard

Schulz

If the community would be willing to fund for better AI, it could be achieved with way cheaper costs actually.

For example, I could contact with developers in my area for this task if the community could determine a budget.

LaFayette

@Schulz dev's cost usually around $60-$100/hr. Any new dev is going to take 10-20 hours just to ramp up on the code before they start doing anything. Bug fixes require anywhere from 1-5 hours each, sometimes more. Small/moderate features will be in the 5-60 hours range.

So, a new dev will be about $1000 just to get up to speed. Each bug is going to cost $500.

I am actively discouraging paying developers, it'll cost a lot. Furthermore, code is an ongoing cost, if they plop stuff down and walk away, what happens if there are bugs when it comes time to release and we roll all the code back because they are not around to fix it? What happens when later we want more tweaks and we are out of budget?

If we can sell dev's to contribute and donate their time, because this is good experience for them and working on AI is interesting, that will be much more valuable. Development labor is generally just too costly. (Perhaps it'll make sense why for the last 4 years I've been focused on trying to make the developers more efficient and not having to spend time on things like map uploads, or why I say that developer time is the most scarce and valuable thing we have for this project).