[Home]GresSuffurd/WeefSuffurd

Robo Home | GresSuffurd | Changes | Preferences | AllPages

Difference (from prior author revision) (major diff, minor diff)

Added: 7a8,16
* 0.2.2 (20070612)
** remove wall-segmentation, current absolute velocity is the only segmentation left
* 0.2.1 (20070607)
** remove segmentation on distance
** re-introduce wall-segmentation (5 segments), take preferred heading (without wallsmoothing) iso actual heading to calculate
* 0.2.0 (20070523)
** segment also on distance
** small preference current direction
** small enhancements virtual waves

Added: 52a62,63
| 0.2.1 | 99.46 | 96.65 | 95.79 | 97.30 | 22.86 | 15.33 | 46.23 | 17.86 | 26.47 | 48.88 | 2 seasons
| 0.2.0 | 99.56 | 96.19 | 96.09 | 97.28 | 18.53 | 13.40 | 49.68 | 18.00 | 27.03 | 47.61 | 2 seasons

Added: 56a68
| 0.2.2 | 99.50 | 96.50 | 96.68 | 97.56 | 13.92 | 8.68 | 40.83 | 14.43 | 21.31 | 44.26 | 2 seasons

Added: 72a85
| 0.2.2 | 98.30 | 92.96 | 91.09 | 94.11 | 24.28 | 15.56 | 46.47 | 21.68 | 27.90 | 48.76 | 30 seasons

Added: 104a118,121

All my tryouts to improve seem to help nothing in the rumble. Now I set 0.2.2 as base for future development and only do FastLearning? MC2K6 for testing. -- GrubbmGait

As base for further development 0.2.2 is significally worse with long battles, but better in FastLearning?, and that is what matters. -- GrubbmGait

GresSuffurd uses a movement that from a (very very large) distance looks like WaveSurfing. The first version (0.0.3) is largely based upon the WaveSurfingTutorial written by Voidious. It will gradually improve until it is good enough for The2000Club.


version history (newest entry on top)

MC2K6 500 rounds

Version WSCBotA WSCBotB WSCBotC WSC APMC CassiusClay FloodHT Shadow CFC Overall Score
0.1.8 99.22 92.72 91.52 94.49 24.21 16.48 48.11 20.64 28.41 49.04 2 seasons
0.2.1 99.46 96.65 95.79 97.30 22.86 15.33 46.23 17.86 26.47 48.88 2 seasons
0.2.0 99.56 96.19 96.09 97.28 18.53 13.40 49.68 18.00 27.03 47.61 2 seasons
0.1.9 99.32 94.23 92.92 95.49 19.84 8.41 53.20 14.83 25.48 46.94 2 seasons
0.1.5 99.60 92.90 92.63 95.04 17.85 7.04 51.47 15.26 24.59 45.83 2 seasons
Grb 89.69 95.88 93.77 93.11 27.77 9.83 16.19 15.11 13.71 44.86 5 seasons
0.1.4 98.16 93.71 91.17 94.35 16.43 8.07 48.54 14.19 23.60 44.79 2 seasons
0.2.2 99.50 96.50 96.68 97.56 13.92 8.68 40.83 14.43 21.31 44.26 2 seasons
0.1.3 99.45 77.99 86.99 88.14 17.19 12.98 46.51 11.37 23.62 42.98 2 seasons
0.1.2 99.18 70.54 89.42 86.38 12.98 12.56 47.92 10.23 23.57 40.97 2 seasons
0.0.6 98.85 50.41 59.65 69.63 12.90 8.46 40.55 8.59 19.2 33.91 2 seasons
0.0.7 99.15 48.70 50.94 66.26 17.80 10.96 30.81 10.94 17.57 33.87 2 seasons
0.0.5 96.95 59.21 48.30 68.15 12.22 9.32 32.61 6.41 16.11 32.16 2 seasons
0.1.0 99.23 63.16 63.69 75.36 4.05 7.94 33.80 4.19 15.31 31.57 2 seasons
0.0.8 99.41 61.33 60.53 73.75 5.99 8.01 31.61 4.37 14.66 31.46 2 seasons
0.0.4 93.70 69.01 64.63 75.78 4.37 4.60 17.20 2.20 8 29.38 2 seasons
BasicSurfer 91.59 58.87 61.09 70.51 6.22 5.46 12.41 3.45 7.10 27.94 2 seasons
0.0.3 74.05 12.98 3.93 30.32 1.85 1.68 4.90 1.57 2.71 11.62 2 seasons

MC2K6 FastLearning

Version WSCBotA WSCBotB WSCBotC WSC APMC CassiusClay FloodHT Shadow CFC Overall Score
0.2.2 98.30 92.96 91.09 94.11 24.28 15.56 46.47 21.68 27.90 48.76 30 seasons
0.1.5 98.83 89.11 87.43 91.79 20.74 12.37 44.85 20.49 25.90 46.14 30 seasons
0.1.2 98.82 70.43 82.71 83.98 16.32 13.26 44.01 13.56 23.61 41.30 30 seasons
0.1.0 98.92 62.29 59.17 73.46 10.13 8.34 36.05 7.69 17.36 33.65 30 seasons


Ouch, I think I still have some work to do . . . -- GrubbmGait

v0.0.4 seems good enough to release in the wild. -- GrubbmGait

The score against WSCBotA is on level, seems like segmentation is the next thing to explore. -- GrubbmGait

It hits the wall a lot, thats something to look into first, together with the right implementation of onDeath. -- GrubbmGait

Still getting the HOT-avoidance optimal and the details right before tuning, read segmenting, to avoid non-HOT. -- GrubbmGait

First time above GrubbmGrb! Against botA, botB and botC the results are acceptable, but I seem very vulnerable for a Bee-like GF-gun. -- GrubbmGait

Checking the details I saw a relatively low score (93%) against bbo.RamboT where I expected a 99%. It turned out that for 3-4 ticks there was no surfable wave, the time it takes a wave to pass my bot. No surfable wave means no movement, which is not good against rambots/close fighters. In movement 0.1.6 (GresSuffurd 0.2.2) this is solved by creating a simulated wave when no surfable wave is available (instead of no wave available). -- GrubbmGait

It turns out that, against the better bots, onBulletHitBullet? happens approx 100-110 times per battle (is 3 times per round), so that information is really welcome. -- GrubbmGait

Wow, that often? Thanks for posting that. =) Now I *really* want to fit that into Komarious... -- Voidious

Some nuancing: the above was with robocode 1.2.5_Alpha. With 1.1.2 it happens 'only' ~70 times (2 per round). Don't know why, have to ask Fnl. -- GrubbmGait

Well, a 3:2 ratio doesn't seem like that big a deal to me... But in 1.2, I seem to be getting zero BulletHitBullets? ever. A sample match of Dookious vs Roland gave 67 in Robocode 1.07 and 0 in 1.2. :-\ -- Voidious

Yeah, I know that the implementation in 1.06/1.07 was not really well (using a line with fixed 8 length I think). In version 1.2.1 (according the version.txt) this has been replaced by a new algorithm. I think that in v1.2 it just was really buggy. I don't have a problem with a better solution, although it might have a minor influence on the performance, as long as it is logical and bugfree. I tested against WaveSerpent with 1.0.6 (59 bullethitbullet), 1.1.2 (63) and 1.2.5_Alpha (106). It seems that currently the most stable/bugfree version is still 1.1.2/1.1.3, but this may change with 1.2.5 (memory issue!) -- GrubbmGait

I like it when things turn out the way they should be: rolling the surfingstats does perform much better against PM and GF-guns, but looses points in the rumble, see 0.1.8. -- GrubbmGait

All my tryouts to improve seem to help nothing in the rumble. Now I set 0.2.2 as base for future development and only do FastLearning? MC2K6 for testing. -- GrubbmGait

As base for further development 0.2.2 is significally worse with long battles, but better in FastLearning?, and that is what matters. -- GrubbmGait


Robo Home | GresSuffurd | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited June 12, 2007 13:17 EST by GrubbmGait (diff)
Search: