Version 12.1

Jump to navigation Jump to search
Revision as of 14 February 2013 at 18:43.
The highlighted comment was created in this revision.

Version 12.1

This version is like a fresh breath of chaos. Oh, the mischief I'm causing. :P

    Skotty00:43, 13 February 2013

    One of the few altruistic bots: beating the strong and helping the weak ;-) But the overall result is ok, closing in on Dookious. If you can find a way to rule out those 98->75 scores you'll be close to #3 and way ahead of me.

      GrubbmGait01:50, 13 February 2013

      I've been impressed with the improvements in GresSuffurd. It has inspired me in my efforts to improve XanderCat.

        Skotty05:11, 13 February 2013
         
        Edited by author.
        Last edit: 20:17, 14 February 2013

        Nice work!

        Though it looks like you still have a few bugs in your specialized modes. A 26% score against GeomancyBS is almost as bad as 70% against Idem (a nano with Linear Targeting).

        UPDATE: Version 12.2 is much better against those two bots.

          Sheldor03:48, 13 February 2013

          Indeed. There is still some room for further improvement if I can iron out those trouble areas without hurting performance elsewhere. I'm pretty sure it can be done.

            Skotty05:05, 13 February 2013
             

            Really nice work, congrats! I wonder if you've got Skilgannon thinking that maybe Bullet Shielding is the gateway to 91 APS - I know I'm thinking it. :-)

              Voidious18:52, 13 February 2013

              Indeed, seeing the diff between GeomancyBS and DrussGT has had me wondering about this for quite a while...

                Skilgannon21:47, 13 February 2013
                 

                Wow... Should I just hold my congrats until you get to #1? =) Friggin' awesome!

                  Voidious07:54, 14 February 2013

                  Thank you. I really didn't expect to see such success with adding a new mode to XanderCat. But after a lot of work unsuccessfully trying to improve the main surfing drive, I have to say I'm most pleased with it. :)

                    Skotty08:13, 14 February 2013
                     

                    I'm also surprised how much it helped. I also think there might be scope for the opportunity to open up a shadow enough to completely hide in it with regular surfing, rather than just have a separate mode. And to get the full magnitude of the improvement, I think this diff is more relevant, and even more impressive. Never mind 91APS, 92 could be a possible target.

                      Skilgannon10:18, 14 February 2013

                      Speaking of bullet shadows, I think active shadowing could be possible in a 1000/1000 field. It could be used in Melee endgame duels to great effect. First, check if a full shield is possible, if it is, then fire a low power bullet to create the shield and move accordingly. If a full shield is not possible, then just surf and fire normally.

                        Sheldor20:22, 14 February 2013
                         
                        Edited by author.
                        Last edit: 20:42, 14 February 2013

                        I'm not sure that's true (edit: active shadowing more effective on larger fields). Being far away means the shadow has had time to grow, that's true. But another key to creating big shadows is for the bullet/wave intersection to happen as soon as possible after the enemy wave is fired. The farther away you are, the harder it will be to do that with any precision. It might be more effective to stay close and create accurate shadows than to gamble from far away.

                        Edit: Sorry, my argument kind of misses the point. With Robocode accel/decel, it's more about giving yourself enough time to get there. So I think you're right.

                          Voidious20:31, 14 February 2013

                          /me suddenly wonders if he's re-tuned his distancing since adding Bullet Shadows... =)

                            Voidious20:37, 14 February 2013
                             

                            I don't understand, I thought as long as there was a 36px area within reach that could not possibly be hit, a bot in that area would be completely safe.

                              Sheldor20:42, 14 February 2013
                               

                              You're right, I was being dumb.

                                Voidious20:43, 14 February 2013
                                 
                                 

                                One bad thing is that there is a lot more volatility in the score, based on whether or not the "scenario" (my terminology) correctly identifies which opponents to activate against. And the score difference can be huge. While in the past the Rumble score was pretty stable after about 2000 battles, I would say it now takes more like 6000 to 8000 battles. Watching the progression on version 12.2, after about 1200 or so battles the APS was around 86.85 (excitingly close to WaveSerpent), but now with 3800 battles it has dropped to 86.50, in danger of dropping below Dookious after such a solid first showing. I still don't think we can say whether or not it will level out above or below Dookious.

                                It should also be noted that the effectiveness of bullet shielding will be less if implemented by top surfers. The score boosts will be less against the vulnerable, and there is more opportunity to lose performance against the non-vulnerable.

                                I also have to take an extra step in future development. Due to the volatility, whenever I am working on anything but bullet shielding, I have to disable the bullet shielding mode to get more reliable testing results.

                                I know I am kind of focusing on the downsides here, but I think the upside is pretty obvious, thus I am commenting on the more subtle issues this new mode introduces.

                                  Skotty19:22, 14 February 2013

                                  After looking at the individual battles against a number of opponents, I might add that with more testing and time, I can probably improve the scenario to activate a little more reliably against vulnerable opponents. This could eliminate at least some of the volatility and possibly give another 0.2 to 0.3 APS boost (assuming 12.2 holds around 86.5). I will have to create a challenge of shielding vulnerable opponents and collect more diagnostics to see which of my conditions for continuing bullet shielding are occasionally violated and why; then I can hopefully tweak it a bit to make detection more reliable.

                                    Skotty19:43, 14 February 2013
                                     

                                    0.35 is a lot after 2k, and I would certainly buy that this increases your volatility, but I'd wait and see if future versions behave similarly to say for sure. I've seen lots of variance above 2k battles with lots of bots, not just after adding sensitive multi-mode stuff. It was something I never really noticed until we had a surge in RoboRumble power from KID's clients and bots started actually getting to 5k-8k battles. It could have also been something like a few enemy bots crashing/skipping turns on one client and then behaving normally on another client that started up later.

                                      Voidious19:38, 14 February 2013