Difference between revisions of "Watermelon/Development"
Jump to navigation
Jump to search
m |
m (→Changelog) |
||
Line 7: | Line 7: | ||
;6a - G1905.6 - same as 6, major bugs fixed, enforces preferred distance | ;6a - G1905.6 - same as 6, major bugs fixed, enforces preferred distance | ||
− | :Fixed a number of silly bugs with version 6. Automatic segmentation should actually provide a benefit now. | + | :Fixed a number of silly bugs with version 6. [[Segmentation/Autoselected Segmentation | Automatic segmentation]] should actually provide a benefit now. |
:Changed distance enforcement to move closer as well as to move further away. | :Changed distance enforcement to move closer as well as to move further away. | ||
:''I suspect more bugs. More testing! | :''I suspect more bugs. More testing! |
Revision as of 05:03, 17 June 2009
Changelog
- 7 - ? - bugfixes, energy management changes, variable tweaking, agonizing
- Altered energy management to take into consideration the recorded accuracy of the chosen segment.
- Decreased bin count and rolling depth on wavesurfing only to better utilize the much smaller amount of available data.
- Altered multi-wave surfing to use the square of the impact-time ratio, which helps decrease the impact of distant waves.
- Changed segment fitness function to use the Crest factor.
- 6a - G1905.6 - same as 6, major bugs fixed, enforces preferred distance
- Fixed a number of silly bugs with version 6. Automatic segmentation should actually provide a benefit now.
- Changed distance enforcement to move closer as well as to move further away.
- I suspect more bugs. More testing!
- 6 - G1900.0 - automatic segmentation, lots of segments
- As promised in version 5, now selects segmentation automatically, inspired by Fractal's strategy at AutomatedSegmentation on the old wiki.
- Modified bin smoothing function to
Math.pow(0.35, Math.abs(index - count))
. - Generally beats version 5.
- Buggy release! This version carefully collected information... and then put it all in the same couple places. Version 6a should perform much better.
- 5 - G1917.7 - multi-segment wavesurfing
- Extended changes for guessfactor gun to wavesurfing.
- Planned for near future: automatic re-segmentation with a large number of factors.[1]
- This release ranks lower than 4, despite the addition of an acceleration segment on the avoidance side. More investigation is merited.
- 4 - G1933.4 - multi-segment guess factors
- Modified guess factors to include acceleration. This has been changed in such a way that adding additional factors later should be straightforward.
- 3a - G1910.0- fixed display of guessfactor waves
- Now shows the wave for the correct segment - sometimes the gun is cool enough to fire but needs to wait another tick for the barrel to swing into position.
- Entered in RoboRumble, ELO 1672.1 after all pairings completed.
- 3 - finished guessfactor gun, increased adaptive bullet power slightly
- Often beats CigaretBH, consistently beats GrubbmGrb, sometimes beats Phoenix (wow! didn't expect that).
- Still doesn't sweat the rambots, even GrubbmThree.
- Next-wave surfing is now weighted by each wave's time-to-arrive and power.
- 2 - full-flight threshold reduced to 80px, also adapts bullet power based on distance
- Performs much better against rambots, and sometimes kills bots with good guns and movement.
- Now surfs the current wave + 1/4 of the next one
- 1 - Version tracking starts here
- Movement: Wave Surfing segmented on velocity / Gun: Linear Targeting (trig).
- MovementChallenge2K6/Wavesurfing: A:99.674, B:90.784, C:94.800 - average 95.086
- Can't beat all the samplebots, mostly because it can't hit them. Consistently, savagely beats RamboT and usually survives against Sanguijuela.