Talk:Anti-Surfer Challenge/Results
Hmm... Voidious, I think it is Diamond, not Dookious? » Nat | Talk » 14:11, 4 August 2009 (UTC)
- Oops... No, these are the Dookious gun results from our pre-vote testing, I just filled in the wrong gun type. Thanks, --Voidious 14:20, 4 August 2009 (UTC)
Wow, +13 against YersiniaPestis? Really?? If that's not a discrepancy in the order of bots in your .rrc file, I am incredibly impressed. =) Nice work! --Voidious 20:57, 10 August 2009 (UTC)
Oh my! This is incredibly impressive! I'm quite curious about what kind of changes led to this incredible result! :) --Rednaxela 21:24, 10 August 2009 (UTC)
Yeah, I thought the high score against YP was weird too -- there are a lot of 99% scores although some dip down to the mid-70s. But it's held up over 50 rounds and I used the .rrc and bots from the zip file so it's not like I'm doing anything different. This is using 1.6.1.4 too. I guess zyx has some work to do?
What's different? This is the version I alluded to on my talk page, using a slightly different set of targeting data and double the number of RBF inputs. Making for a really large number of NN weights. I've hesitated using this targeting in Holden, since the cpu usage may become too high when combined with wave surfing. On the TC2K7 this gun scores about 0.5 higher than the posted score for Gaff; it still can't break 88 in the TCRM though.
--Darkcanuck 21:53, 10 August 2009 (UTC)
Nice work, Voidious! Competition is good... =) --Darkcanuck 21:55, 8 September 2009 (UTC)
- Thanks! I couldn't be happier with big boosts in both PL and APS. An unbiased VG got 72.51, but my AS gun alone got only 71.71. The final 1.34 is a VG with a heavy bias against the AS gun at medium to high hit percentages. Looks like we've got even more competition from Navajo now, too. =) Nice job, dude! --Voidious 23:20, 8 September 2009 (UTC)
Wow, nice scores Navajo! There is such a thing as too much competition, you know... Will Gibbs be entering the rumble anytime soon? --Darkcanuck 23:24, 8 September 2009 (UTC)
- It deppends on what do you mean by soon, but I think until April next year I will enter it on the rumble, its movement still needs a lot of work, and it still isn't that good at TCRM. Also, I have some interesting ideas I'm willing to try before releasing it. As a side note, the DC part from 1.160 is a very simple one I worte, debugged and optimized in something like 2 hours using Rednaxela's kd tree, while the 1.170cDC is a version using one single pure DC gun alone very optimized. However, when I tried to implement this last DC gun together with the rest of the bot the overall preformance decreased, and a deeper analysis revealed it was actually worse, and its results were better because it caused the bots to activate their flatteners later... --Navajo 23:44, 8 September 2009 (UTC)
