[Home]ZoomTargeting/Stage1

Robo Home | ZoomTargeting | Changes | Preferences | AllPages

Difference (from prior major revision) (no other diffs)

Added: 34a35,36
* yes .. that is strange .. it sometimes gets the weakness, but not nearly as good as DT does .. I think that it is because Tityus moves between segments too quickly. I'll have to think how (and if) i should solve this -- Vic


Added: 39a42,43
* Currently the MacroSegments? don't yet have statistical properties, so they are much less accurate then the MicroSegments. The virtual gun type approach is in fact also part of the design. Each segment will have a hit ratio and when aiming I will choose the segment with the highest hit rate. In the next stage I will implement this. -- Vic


Added: 44a49,50
* I already posted the results for Mazer, which has the ZoomTargeting gun WITH MultipleChoice stats in the MacroSegments?. I'll post results for PatternMatcherBot also. --Vic


developement stage 1:

firing waves:

aiming:

After thorough testing of the properties of this gun I realise that in this stage it's not yet competitive gun against the state-of-the-art guns. You'll find that it loses to the top bots. It's about equally effective as normal PatternMatching guns. (So not bad...)

good things: - it learns very fast.. in early rounds of a match (no saved data) it beats most other top statistical guns. I hope i can retain this property when adding MultipleChoice in /Stage2?. - its performance is very fast (maybe it could be on the list FastBots?;-). - it punishes the enemy if its movement is not flat in short timeframes.

bad things: - it learnes slower than top stat guns in later rounds, because it is lossy in the lower segments. - it still uses a lot of memory, although it is now capable of running 1000 rounds without crashing my system :-). - it cannot punish the enemy if its long term movement profile is flawed (not flat), but its short term movement profile is flat.


About its ability to find non-flat movement in short time frames. Only in certain conditions I think. Since otherwise it would have spotted the weakness of the Tityus 0.2-0.3 movement. You should probably keep that version of the movement just for testing purposes. I know I will. =)

When it comes to finding weaknesses in the long term profile. If it's really there you should be able to zoom out, shouldn't you? Maybe if you could use a virtual guns type of approach with different zoom levels? I don't really understand the theory behind zoom targeting so I might be way off here.

Can't you post your TargetingChallenge results with ZoomTargeting and your pattern matcher for reference? Especially the latter one would be interesting to keep there since it is one of the guns in the MovementChallenge.

PEZ


Robo Home | ZoomTargeting | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited August 20, 2003 19:44 EST by Vic Stewart (diff)
Search: