The new adaptive AI

Discussion in 'General Discussion' started by Paradox Agi, Apr 18, 2016.

  1. clevername

    clevername Member

    Joined:
    Jul 19, 2015
    Ratings:
    +14 / 0 / -0
    Good point !
     
  2. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    Hmmmm....
    Well, in theory one should only need the AI lap-time for one specific AI level, and then you should be able to adjust from there.
    Example: if the 100% AI runs a car/track combo in 80 seconds and you do it in 73, you should be able to start around level 109-110.
    80/73=109,5
    Again, in theory. :D
     
    • Like Like x 1
  3. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    Yep, confirmed.
    That actually works rather well.
    100% AI laptime divided with my own yielded 101,8 so I set it 102 and got a good fight with no new index written.
    So if you're unsure of your level, run one lap with 100% AI to get a lap-time.
    After that you can simply divide and conquer (if you'll pardon the atrocious pun :D )
     
    • Like Like x 2
  4. digitizer

    digitizer Well-Known Member

    Joined:
    Mar 26, 2016
    Ratings:
    +117 / 0 / -0
    Great info. Tomorrow i'l begin to record AI lap times for all tracks and cars i have :)
     
    • Like Like x 1
  5. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    Don't forget to include the AI level as well. Otherwise the info will be useless.
    Come to think of it, the easiest way would be to simply upload a copy of your adaptation-file.
    Here's mine: http://www.mediafire.com/download/klledr99jprfu4z/aiadaptation.xml

    @mr_belowski you stated earlier that your adaptation-file was full of stuff.
    Any chance for a copy of it?
     
  6. mr_belowski

    mr_belowski Well-Known Member Beta tester

    Joined:
    Apr 23, 2015
    Ratings:
    +1,307 / 0 / -0
    Sure, but it'll be tomorrow evening now.

    Had a few thoughts about this process too, which I'll arrange in a sane order and write down :)
     
  7. pixeljetstream

    pixeljetstream Well-Known Member Beta tester

    Joined:
    Jan 29, 2015
    Ratings:
    +412 / 0 / -0
    I can help parsing the stuff with some scripts and create a database file.
     
  8. jimortality

    jimortality Active Member

    Joined:
    Jun 11, 2015
    Ratings:
    +39 / 0 / -0
    So we don't need to train the ai now?
     
  9. pixeljetstream

    pixeljetstream Well-Known Member Beta tester

    Joined:
    Jan 29, 2015
    Ratings:
    +412 / 0 / -0
    Well the data doesn't come from nothing, ideally we can quickly generate something that makes "priming" files quicker.
    You'd still need to know what lap times you typically do, per class per track

    Think of a tool that let's you enter track and class and desired target time + variance. Then generates the adaptive file for you.

    The tool works better the more people provide their current adaptive files.
     
  10. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    I should probably try that at some point. :D
    Well, we do. What we're trying to do is figure out an automatic way to get it started at the right level.
    The larger your variance in your lap-times, the more levels you're going to need in the XML file.
     
  11. clevername

    clevername Member

    Joined:
    Jul 19, 2015
    Ratings:
    +14 / 0 / -0
    Great thread! Question... what is the effect of off-track/invalid laps on the competitiveness of the AI. If I have, say 2 invalid laps out of 10, and the 8 valid laps are consistent. How much will that affect the AI in the next race?
     
  12. digitizer

    digitizer Well-Known Member

    Joined:
    Mar 26, 2016
    Ratings:
    +117 / 0 / -0
  13. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    You could, but if you're going to run one lap anyway you might as well do it with AI and get the index written automatically.
     
  14. Supermini

    Supermini Member

    Joined:
    Mar 31, 2016
    Ratings:
    +21 / 0 / -0
    Just wanted to say that I tried @CheerfullyInsane 's AI training last night, and it worked perfectly. Did a bit of a qualifying session against a 100% AI to get my AI level (was down by -0.5s, so 100% :p ), did my six laps to get the 98-103 indices, then started a 20-car race against the Adaptive AI. Started P10, worked my way up to P4, was nose-to-tail with P3 for a couple of laps trying to pass, then spun on the penultimate lap, finally recovering to P8. Great race!

    It seems to me that an AI training mode could/should be part of the game. For one thing it would clear up new player confusion about adaptive AI seemingly not being any use. I understand priorities and resources and rejiggering the menus and testing all that, but conceptually at least it's pretty straightforward. Step 1, have the player pick a car and track, then go out and do as many laps as they like against a ghost AI at 100%. When the player has a lap time they're satisfied with, they can continue to step 2, six laps against a ghost AI at (AI's step 1 lap time / player's step 1 lap time) - 2%... etc. Done! I am not a developer, but that doesn't seem too infeasible.

    And they can add Apex Hunt back while they're at it. :D
     
    • Agree Agree x 1
  15. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    Could be done a lot easier.
    Simply have the AI write an index for every lap, instead of after each race. That way you'd get a working index after one 30min race.
    But then, I'm not a programmer either so what do I know? :D
     
  16. pixeljetstream

    pixeljetstream Well-Known Member Beta tester

    Joined:
    Jan 29, 2015
    Ratings:
    +412 / 0 / -0
  17. Skybird

    Skybird Well-Known Member

    Joined:
    Mar 31, 2015
    Ratings:
    +803 / 0 / -0
    Hm, following this thread since days, I decided to give the adaptive AI a second try. Its long time ago I tested this feature, and it was when I had far less hours with raceroom under the belt.

    Honestly said, something has changed - for the better. I do not kn ow whether it is the feature itself, that they have tweaked it, or if it simply is me and my perception now, today, but having had several 10 minutes races with randomised time of day and starting positions and always default car setups, no training and no quali sessions, after just three races had me racing against GT3 cars, later also GTO cars and DTM2015 cars, that indeed had nicely adapted to my own skill level. It works much better than I recall it did last year, in summer, or whenever it was.

    So I am happy to admit that I cannot hold up my former opinion anymore that the feature did not work too useful. It does, and I really like it now. Whether it is a change in me, or a change in the code, I cannot say.

    At least my needs are well served with it.

    I'd like to know, however, how it works if you set up a championship. Will the AI still make small changes from race to race, or will it use one and the same AI setting found by the adaptive AI at season start for the whole season?

    If it is longer time ago, many months, you tested the feature yourself in SP, give it another try, maybe, just two or three short races may be enough to make things klick right for you. Just make sure you do not mess up your result by crashing, spinning off, but do a clean race and finish.

    My apologies to S3 if I ever criticised this feature in the past in harsh words, while the adaptive aI ha dnot been changed in this time indeed. I do not remember that I ever became harsh in words over this feature, but memories may have escaped me.

    P.S. Before, I used fixed AI at 100, 110 or 120% (with default car settings), depending on car class.
     
    • Informative Informative x 1
    Last edited: Jun 4, 2016
  18. maxx2504

    maxx2504 Member

    Joined:
    May 3, 2016
    Ratings:
    +20 / 0 / -0
    Hi !

    I spend a lot of time to find this file but i can' find it ? I drove with Adaptive AI a lot of time.

    Whats wrong with my game ?...
     
  19. CheerfullyInsane

    CheerfullyInsane Well-Known Member

    Joined:
    Oct 24, 2015
    Ratings:
    +402 / 0 / -0
    Sorry for the late reply, I must've missed that one.
    Anyway, championships work exactly the same as normal races when it comes to the adaptive AI.
    It doesn't set an AI level for the championship as a whole, it simply uses the indexes already saved for each car/track combo.
    So if you're at around 98 on one track, it'll use that, and maybe switch to 109 on the next track if you're better at that particular piece of real estate.
    The upside is that you don't have to be equally good on all tracks (and I don't know any who are).
    The downside of course is that you have to train the AI for each track before starting the championship.
     
  20. Winzarten

    Winzarten Well-Known Member

    Joined:
    Jun 15, 2016
    Ratings:
    +377 / 0 / -0
    This is a very informative topic, thanks a lot guys for the research :)

    Just a simple question from a new player. In first time races with a car/track combo, what does the adaptive AI use as baseline for difficulty?
    Does it have some value set, or is it trying to estimate som kind of avarage AI difficulty based on other tracks/cars?

    I'm asking because I started playing yesterday and selected adaptive AI.
    The first race, at the raceroom racetrack, was easy and beat the AI by a large margin (40+ second after doing a non-mandatory pit stop), according to the .xml file the AI strength was 90
    I then moved to slovakia ring and was around -2 seconds off the pace from the lead. The AI was at 102, so I'm wondering how it ended up with that number? Was is a random from an interval, or it looked at the previous result and adjusted according to that?