Difference between revisions of "Glacier/GlacialHawk"
< Glacier
Jump to navigation
Jump to search
(utter failures) |
(GlacialHawk 1.8) |
||
| Line 8: | Line 8: | ||
<!-- {{subst:rumblestats:meleerumble|ags.muse.GlacialHawk x.x|RumbleStatsDefault}} --> | <!-- {{subst:rumblestats:meleerumble|ags.muse.GlacialHawk x.x|RumbleStatsDefault}} --> | ||
| + | * '''1.8''' - 22/9/2009 | ||
| + | ** ''Rating Pending'' | ||
| + | ** Revert to 1.5 | ||
| + | ** Tweak bulletpower very slightly smaller | ||
| + | ** Instead of always predicting 12 futures per enemy, predict up to 30 per enemy, with a maximum of 100 predictions in total for all enemies | ||
| + | |||
* '''1.7''' - 20/9/2009 | * '''1.7''' - 20/9/2009 | ||
** Rating: ''Utter failure...'' | ** Rating: ''Utter failure...'' | ||
| Line 15: | Line 21: | ||
* '''1.6''' - 20/9/2009 | * '''1.6''' - 20/9/2009 | ||
| − | ** | + | ** Result: Slightly worse than 1.5. RumbleStats doesn't represent non-current bots well so not bothering to post rating. |
| − | |||
** Only put results into the tree after 90 ticks. | ** Only put results into the tree after 90 ticks. | ||
* '''1.5''' - 20/9/2009 | * '''1.5''' - 20/9/2009 | ||
| − | ** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=meleerumble&name=ags.muse.GlacialHawk+1.5|rumble=MeleeRumble|scorelabel=APS|score=68.37|rank=6th|win=278|loss=5|plrank= | + | ** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=meleerumble&name=ags.muse.GlacialHawk+1.5|rumble=MeleeRumble|scorelabel=APS|score=68.37|rank=6th|win=278|loss=5|plrank=4th|glicko2=1646.4|score2label=Survival|score2=88.89}} |
** Revert to 1.2 (for now) | ** Revert to 1.2 (for now) | ||
** Increase the number of 'proximity sensor' dimensions and tweak dimension weightings | ** Increase the number of 'proximity sensor' dimensions and tweak dimension weightings | ||
Revision as of 03:45, 23 September 2009
- Glacier Sub-pages:
- Glacier - Version History - GlacialHawk
- 1.8 - 22/9/2009
- Rating Pending
- Revert to 1.5
- Tweak bulletpower very slightly smaller
- Instead of always predicting 12 futures per enemy, predict up to 30 per enemy, with a maximum of 100 predictions in total for all enemies
- 1.7 - 20/9/2009
- Rating: Utter failure...
- Early results: MeleeRumble ‒ APS: 68.19% (5th), PL: 277-5 (5th), Survival: 88.74%
- Revert to 1.5
- Bias against returned situations which are more dissimilar.
- Rating: Utter failure...
- 1.6 - 20/9/2009
- Result: Slightly worse than 1.5. RumbleStats doesn't represent non-current bots well so not bothering to post rating.
- Only put results into the tree after 90 ticks.
- 1.5 - 20/9/2009
- MeleeRumble ‒ APS: 68.37% (6th), PL: 278-5 (4th), Survival: 88.89%
- Revert to 1.2 (for now)
- Increase the number of 'proximity sensor' dimensions and tweak dimension weightings
- 1.4 - 19/9/2009
- MeleeRumble ‒ APS: 68.22% (4th), PL: 278-5 (4th), Survival: 88.37%
- Tweak the new probabilistic model introduced in 1.3, to include a 'doubt' factor based partially on distance. (Maybe make the 'doubt' factor learned in a future version?)
- 1.3 - 19/9/2009
- MeleeRumble ‒ APS: 67.65% (5th), PL: 274-8 (9th), Survival: 87.23%
- Fancy new experimental modeling of bullet-hit-probability at possible angles. Should work better than simply summing the movement profiles in the common 'shadow's melee gun' style. At least, it should once I get it tweaked...
- 1.2 - 18/9/2009
- MeleeRumble ‒ APS: 68.39% (4th), PL: 276-5 (6th), Survival: 88.72%
- Large improvement in PIF speed, at the cost of reduced wall-intersect checking.
- 1.1 - 10/9/2009
- MeleeRumble ‒ APS: 67.72% (5th), PL: 276-5 (5th), Survival: 87.82%
- Increase bullet power notably, to mostly follow HawkOnFire bulletpower formula
- Huge SlowBot and skips occasional turns. Will try to fix in 1.2 :(
- 1.0 - 9/9/2009
- MeleeRumble ‒ APS: 67.48% (5th), PL: 275-6 (6th), Survival: 91%
- Movement of HawkOnFire (from here) with Hypothermia (the gun of Glacier).
- Thoughts: Seems to be scoring how I expected: Above DiamondHawk but below ShadowHawk. But considering this gun was only given a single test run before release and no tuning at all, this is probably quite a good rating. Higher survival score than ShadowHawk indicates that bulletpower could be increased to trade a little survival for score, which doesn't surprise me given that the bulletpower selection is a complete 2-second hack without any real thought or tuning.
