Hi, given the excellent analysis by @CheerfullyInsane in this thread https://forum.sector3studios.com/index.php?threads/the-new-adaptive-ai.5013 the idea of this tool was to create a database where people just enter the laptime they do for a certain track/car combo, and a tool generates adaptive files for them that have a user defined spacing (say your medium AI level is 90, and you want AI to be 85-95 or so). That way in theory less training needs to be done by individuals. This works, however due to R3E's AI changing a good deal, it doesn't make the results very stable Nevertheless the tool can be found here, and may still be useful for your needs. https://github.com/pixeljetstream/r3e-adaptive-ai-primer You may want to look at this thread https://forum.sector3studios.com/index.php?threads/adaptive-ai-the-tool-how-to.12888/ how to use the tool WARNING: When applying modifcations, it will overwrite whatever ai levels you had for that particular track/class combo. UPDATE The tool can also be used to reset the AI levels (not shown below, but there is a new button), which is particularly useful when R3E gets major AI changes. The tool also shows your own best lap times that the aiadaptation file is aware of.
Hi ! I searched my whole PC for that file. Can't be found. Hided files are searchable. I don't know why. Maybe i should try to copy it from another user ?
You did actually do a race against the AI, right? It doesn't get generated until you do. But here's mine: http://www.mediafire.com/download/9pkk5a7dp7bgnzj/aiadaptation.xml Not a whole lot in it thus far, been too busy with RaceDepartment stuff to do much off-line racing. I'll update the link as and when I add more stuff to it. Mainly going to be Aquilas, FRJs and other low-power stuff. Everyone and their sister will be doing GT3s anyway, so.... @jimortality @Christian Göpfert @mr_belowski @kendoslow Care to chip in?
@pixeljetstream Something just occured to me. If you get this up and running, can you make it so it lists both the AI levels and the approximate lap-times it'll do? That way we can simply hack the XML, and bypass the training runs altogether. (At least in theory..... )
Mine is here https://forum.sector3studios.com/index.php?threads/the-new-adaptive-ai.5013/page-6#post-72522
Sorry to say I have nothing to contribute as I have never ran a single race against the AI. Maybe time to give it a go, but when I do fancy an offline race I tend to fire up Race07 or GTR2.
Gonna ask if it's ok to share mine (as those are always also used in internal). Should be, but just in case.
exactly, the idea is that the tool writes the XML for you you pick the target level based on the lap-time you did + how much variance we should prime the xml file with. @Christian Göpfert you can also pm me the file and I'll rip out hints to unreleased content (can sign NDA if I have to as well). The way the files are organized it's not visible what cars/tracks, just the ids, but yes you may have ids that are not publicly documented. You could search for those unpublished ids and delete the sections in the file prior sending as well. The class ids I use are from the portal html code, see: https://github.com/pixeljetstream/r3e-open-championship/blob/master/assets.txt
Alright thanks guys, we have something http://htmlpreview.github.io/?https...aptive-ai-primer/blob/master/docs/sample.html No actual tool yet, but at least some database view... The best way to handle this would be webinterface where people send in files, and the database is always updated... but I am not a web programmer Next step is a ui frontend that modifies your aiadapatation file based on the database and what your target AI levels would be. However as the numbers currently show, we would need tons more files, to get reliable numbers.
looking into the files a bit deeper, it worries me that the variance is so high. So all the timing values where it's not just 0.0000 (single value) we have variances that go up to 2 seconds. I also saw some inversion, like lower AI faster than higher AI because different people drove. That would kinda suck, if AI performance is very machine dependent. Obviously it's gonna be flaky a bit, say one lap had a car crash in it... but still we will see how it works out. If not, well theory at least was tested s3s could generate the numbers by running the AI simulation over every track/car for different AI levels (no visuals loaded, so could be computed quickly).
I just had a look, watch out for the Nords times, i know nothing of coding n shit but however you are reading them the dis[play is wrong, maybe im reading that list wrong but certain times appear to be missing the 0, 7:1.3252 7:5.0966 7:2.6565 7:1.1512 Bathurst 2:7.1534 thats not 2 mins 70 Maybe im reading it wrong like i say.
I'll get onto RD, and see if I can't convince some of our regular club racers to pitch in. Annoying, but not that big of a problem. We are after all only looking for a rough estimate, with 2-3 levels of AI on either side. If one or two entries are a little off, you should still be in the right ballpark.
@pixeljetstream hi,i only got 5 or 6 Races against the AI.Is it worth anyway for the Database?I will do some Race's against the AI till mid next Week for testing Tyres/Fuel and Pitstop strategies over longer runs.Maybe should upload then my aiadaptation.xml file.