Forged Alliance Forever

Re: AI update

2017-02-16T22:16:31+02:00

Indeed - platoon composition was always a problem for the original and Sorian AI. LOUD takes platoon composition in a direction that takes advantage of the 'squads' features within the existing platoon code - and has more specific platoons for certain purposes. Alongside getting a better handle on the categorizations of the units themselves, it does make it possible to get some marked improvement in how the AI uses, and builds units.

Statistics: Posted by Sprouto — 16 Feb 2017, 22:16

Re: AI update

2017-02-16T21:10:05+02:00

Part of the issue i saw with the AI defined platoons is understanding what each one did in relation to the unit types provided. Often were more the one group-type a unit could be lumped into just based off the rather sparce description provided or what was known about each platoon based on the units currently listed. Having verbos platoon descriptions / documentation would help future modders understand this process without all of the trial and error i went thru.

Beyond this... adding a custom unit to a platoon, would often prevent the standard unit from being built. I.E. Swapping one unit for another ect. This is one of the things i saw as being a major limitation of the platoons and one that could have serious repercussions if the wrong units were swapped out. (Imagine if no engineers were built)

Instead, I'd much rather see 3-5 Custom platoon types be made avalible with controls on how many, how often, what tier and what for. Unfortunately as i understand it this isn't possible without upsetting the current AI build patterns.

Statistics: Posted by Resin_Smoker — 16 Feb 2017, 21:10

Re: AI update

2017-02-16T18:43:10+02:00

It's not magic - it's merely an extension of the Sorian AI method. LOUD adds additional category definitions to those that were supported by Sorian so that it can better understand how units are to be used. This is of particular note for amphibious units which were often built as part of the standard unit choices (particularily in T3 where the original definitions had no amphibious choice at all). In those cases where a custom unit is categorized along with existing units, the custom unit file can specify a percentage chance that the AI will select that unit rather than an existing one.

For experimentals, this idea has been greatly expanded. There are now 4 levels for Land, 2 for Air and 2 for Naval. Artillery and Economic Experimentals are handled independently. The levels are primarily determined by cost, so that the AI can choose one that suits his current eco.

Statistics: Posted by Sprouto — 16 Feb 2017, 18:43

Re: AI update

2017-02-16T13:55:13+02:00

So how is the AI making use of custom units ?

Statistics: Posted by Resin_Smoker — 16 Feb 2017, 13:55

Re: AI update

2017-02-09T14:36:26+02:00

That would be nice indeed.

Statistics: Posted by Krapougnak — 09 Feb 2017, 14:36

Re: AI update

2017-02-07T21:28:02+02:00

That's a good question - HtKatzmarek brought the unit package to me for inclusion with LOUD so I'll have to ask him.

Statistics: Posted by Sprouto — 07 Feb 2017, 21:28

Re: AI update

2017-02-07T12:40:37+02:00

Hi Sprouto,

I was wondering how you managed to port SC2 units into FA under Windows 10.

I did port some SC2 units under Windows XP but since I moved to W10 the program I used to convert animations doesn't work anymore.

Statistics: Posted by Krapougnak — 07 Feb 2017, 12:40

Re: AI update

2017-01-08T10:37:09+02:00

Without knowing what those changes are, I really cannot say.

LOUD features a VERY long list of balance tweaks, many similar to what FAF has recently introduced, but likely far more extensive. In order to get the AI to function better, it was necessary to review the threat values assigned to EVERY unit in the game, using an algorithm the takes many factors into account, with some variation for economic structures, intelligence structures and so on. This process really made it clear that not only were the original threat values way out of whack, but it highlighted a great deal of inconsistencies between unit classes and unit tiers. When that was done, something truly good happened to the AI -- he got smarter, since he was now able to make good decisions about when and where to attack, and with what.

So, long and short, yes -- it would be possible to bring the FAF unit database into LOUD and take advantage of the performance improvements that LOUD offers, but it would likely nerf the AI rather severely.

As for functionality, I'm sure that specific game functions could be incorporated, but the true value of FAF lies in it's community and PvP focus. The FAF lobby, ladder and ranking features are far beyond anything I ever did with the core code for Supcom.

Statistics: Posted by Sprouto — 08 Jan 2017, 10:37

Re: AI update

2017-01-07T22:52:23+02:00

Would it be possible to integrate the current faf balance and functionality into LOUD?

Statistics: Posted by SpoCk0nd0pe — 07 Jan 2017, 22:52

Re: AI update

2017-01-02T00:28:31+02:00

Yes ...I've always hoped there would be more interest, but I'm not really a good publicist, and getting the word out has been difficult. I finally screwed up the backbone to ask my community members if anyone was interested in helping out. To that end, I will shortly post my own thread on LOUD with a demonstration video and a deeper explanation of all that it is and can do. I'll post on this thread with a link when it's up.

I would love to say that yes ! it would be easy to integrate all these advantages into FAF - but it would be white lie. In 3+ years of this I've gone in and re-coded or deeply edited almost the entire exposed code base. If there were just a few key tweaks that made this possible I would gladly do so - but no - there are hundreds, possibly thousands, as is often the case with optimization. Likewise, file structures have been altered too - which is why we have an installer/updater program that will allow users to keep their existing FAF configuration AND a LOUD installation - switching as needed.

I hope that doesn't deter anyone. The LOUD project was begun with the desire of addressing the less than stellar AI of the original game. Obviously, game performance and the AI were closely intertwined. And along the way it became clear that meant units had to be edited, map markers had to be redone, etc.

Statistics: Posted by Sprouto — 02 Jan 2017, 00:28

Re: AI update

2017-01-01T18:08:36+02:00

Sprouto wrote:
I'll try to be short.

It started more than 3 years ago - firstly to just make the AI better.

Along the way, it was clear that the performance issues in the game, and the AI were all intertwined. There were thousands of tiny optimizations and clean ups that resulted in a much tighter core code -- not just for the AI but the entire game. Lots of extraneous (unused) code and data was removed and reworked. Many of the units had either wrong or unused categorizations that needed to be corrected or removed, or were able to benefit from simplifications in their blueprints. That in turn led to an examination of the maps and how the AI interacts with the markers, and the pathfinding issues for the AI, which in turn led to re-marking most of the maps that come with LOUD. Long story short, in the end, almost everything got looked at. That's essentially why it's not just a mod, but a complete stand-alone package.

Now, without too much effort, you really can play a 5000 unit game on 40km maps and not have to worry about slowdowns against an AI, that, on many days, will really put you to the test. Is it a personal effort ? Mostly - and yes, you will find lots of changes in the game, slight differences in the units and structures, and carefully selected pieces of our favorite mods to incorporate into it. We're still looking for more - and obviously some practical input from others - but it's pretty much mature now. We're convinced it's not only a better AI, but a much better performing game, with a much tighter balance that supports all styles of gameplay.

It's not aimed at the 5km or 10km short game, but for the larger longer epic games where anything and everything can come into play. Try it out. Please.

This sounds amazing!
Could we have most of those performance improvements for FAF without breaking things? I mean this sounds like code improvements on a scale that really should stir up much more interest then it seems to do.

Statistics: Posted by SpoCk0nd0pe — 01 Jan 2017, 18:08

Re: AI update

2016-12-05T18:59:32+02:00

We've dealt with the Steam compatability as LOUD will now detect your original installation and save it as is, and will give you the option to switch back and forth as required. This works as well with FAF we are told, permitting users to utilize both as required.

We have had minor issues with Windows 10 taking ownership of the directory structure and not allowing users (even the administrator) to make changes, but it's a rare thing and we have a solution for it if required.

Statistics: Posted by Sprouto — 05 Dec 2016, 18:59

Re: AI update

2016-12-05T18:12:21+02:00

well as it stands a surprisingly large portion of the faf playerbase does like to play with ai, and with mods like blackops, ect.
so i think this could be a pretty nice improvement in that area at least.

things like map markers i think is a bit much, but not totally outside the realm of possibility to put in

after all faf is a kinda overhaul of supcom so either the changes you made are most likely able to be put in since faf does aim to improve whatever it can.

https://github.com/FAForever/fa
why not try making a pr and lets see where this takes us?

im not very familiar with ai code so i dunno how much work it would be, but im sure that we would be glad to have it, along with anything else that you might offer.

Statistics: Posted by Exotic_Retard — 05 Dec 2016, 18:12

Re: AI update

2016-12-05T17:42:37+02:00

Thanks Sprouto! I'll give this a shot tonight.

Last time I messed around with Loud AI (2 years ago) it was not compatible with Steam. Same deal?

Statistics: Posted by MandeR — 05 Dec 2016, 17:42

Re: AI update

2016-12-05T16:12:25+02:00

If you've read the thread you can tell how sideways this project went over time. Eventually it encompassed almost every aspect of the code -- not just the AI. In that regard, it would be a rather large effort to bring all the benefits into FAF, if they're even wanted. However, having said that, much of the structure of the AI would be portable, if not as a replacement, but as an alternative to Sorian, to whom we all owe credit to.

I'm not sure how valuable the AI is to the FAF community, since most of the focus seems to be upon PvP, but I have always hoped that more folks would take an interest in this work. I don't know if anyone is handling the AI code base for FAF but I would be glad to contribute in that area alongside anyone else willing to participate.

Statistics: Posted by Sprouto — 05 Dec 2016, 16:12