Interesting... 1.4.5 beat DuelistMicro by a small margin in the long run, while 1.4.6 loses quite badly... --David Alves
It's perfectly possible it happens. I tested the new version with short battles, to tune it for the Eternal Rumble and Minibot Challenge. -- Albert
About Marshmallow: MicroAspid is one of the few (maybe the only one) of the top bots that is not using VB or "the projected gun bearing to hit" for aiming. Probably Marshmallow has a movement that tries to create a uniform distribution function for this variable. If it does it by alternating certain patterns, it will be of little help to avoid being hit by MicroAspid :-) -- Albert
Well, it's not dodging that is the hardest against MicroAspid. It's hitting it that Marshmallow has most problems with. Marshmallow has an easier time hitting SandboxDT actually. None of Marshmallow's VirtualGuns seems to fit MicroAspid's movement pattern. It takes about 1500 rounds before Marshmallow starts winning more rounds than it loses. -- PEZ
Now, this is one competitive bot! It would benefit greatly from storing data persistantly I think. Make a MacroAspid? please. =) -- PEZ
Uh, there already is a larger version of MicroAspid... it's called "Aspid"... --David Alves
Yes, but it's not as competetive as a MicroAspid with persistant storage would be. My point was that I fear that it's the "Micro" that keeps this bot from becoming what it could be. -- PEZ
Persistant storage doesn't work very well with pattern analysis - storing the patterns takes up a lot of room. --David Alves
Well, it is not completely true. I have been doing some tests, and using the zip compression posted in the Wiki (See CompressedSerialization) it takes 4 to 7K to store data for a pattern matcher. The only problem is that it only gives you advantage for the first rounds (10 to 15). Any way, I'm working in the new Aspid... -- Albert
Since EternalRumble runs on 10 round matches that could make a huge difference there at least. Marshmallow doesn't save data for it's pattern matcher though since it's such a simple matcher that it only takes half a round to recreate it. But for the data it saves I have found that the "transient" keyword can save half the space even with CompressedSerialization. Which btw might not be the best choice for saving pattern matcher data. I think just saving the data (compressed of course) would give you smaller files. Though it's more programmer work of course. -- PEZ
![[Home]](/images/RoboWiki.png)