http://robowiki.net/w/index.php?title=Talk:Pugilist/Version_History&feed=atom&action=historyTalk:Pugilist/Version History - Revision history2024-03-29T06:24:32ZRevision history for this page on the wikiMediaWiki 1.34.1http://robowiki.net/w/index.php?title=Talk:Pugilist/Version_History&diff=7845&oldid=prevVoidious: migrating talk2009-05-29T19:36:13Z<p>migrating talk</p>
<p><b>New page</b></p><div>So the increased granularity of the 1.6.4.1 gun segmentation slows learning down a bit too much it seems. But it makes for a pretty good PremierLeague bot since this segmentation seems strong against the more sophisticated movers. -- [[PEZ]]<br />
<br />
According to the TargetingChallenge/Results I have a badly broken gun in Pugilist. I must check the results though, since 1973 rating points with a 87.59 TC gun sounds pretty incredible. But if it holds true I have a point or two to collect by fixing the gun. =) -- [[PEZ]]<br />
<br />
Well, it seems that it is the smoothing of the gun stats that "breaks" the gun. Now I have an interesting case where two versions of P have about the same rumble rating (1973 and 1972, which is within the margin of error) but dramatically different gun ratings (87.59 and 92.06 respectively)... I know of only one difference in movement in these two bots; the number of stat bins for the enemy waves. Either the TC gun performance doesn't matter at all or the movement is much better with more stat bins. I might let an interim version of P become a megabot just to check... (It costs me bytes to use different number of bins for gun and movement.) -- [[PEZ]]<br />
<br />
Forget versions 1.6.4.15-16. They shared a huge smoothing bug. I'll try those configs later probably. -- [[PEZ]]<br />
<br />
Funny. With 1.6.4.34 I tried to make P stronger against the weak, scrificing strength against the strong. But it came out the other way around. -- [[PEZ]]<br />
<br />
Not so funny. It's rather depressing that all my tries with purer surfing (the 1.9 line) so fails to collect even close to the points the wall-bouncing 1.6 line does.... And frustrating to! -- [[PEZ]]<br />
<br />
Ah! My new 1.9 approach together with [[Axe]]'s good advice to allow wall bouncing at really close ranges made 1.9.3.3 jump up to compete with the stupid 1.6.4 line. Nice! I don't think the 2K club had a slot for 1.6.4.*. But I'm pretty sure a chair is free there for 1.9.3 and up. -- [[PEZ]]<br />
<br />
Gah! I just realized I have a quite huge bug in the surfing of 1.9.3.3. Wonder if I dare fix it. =) -- [[PEZ]]<br />
<br />
Ha! 1.9.3.3 finally managed to grab #1 in minirumble!!!!!!!!!!!!! That was hard work. Now P is the highest ranked mini in general rumble, #1 in minirumble and doesn't lose against one single minibot. Maybe a candidate for mini TheBestBot? It's more fun to say yes! =) -- [[PEZ]]<br />
<br />
Hey, that 1.9.4.3b seems very strong (1990+ with 500 battles!), it´s a gun improvement or movement? -- [[Axe]]<br />
<br />
Strongest Pugilist ever! =) It was movement segmentation changes. Lost a lot against some bots, but gained more over all. -- [[PEZ]]<br />
<br />
Seems that 1.9.5 solved that DevilFish problem:<br />
|Marcelo.Alpha.DevilFISH_1.3 |92.8 |2| 16-5-2004:2:32 |83.2| 9.5 | <br />
What do u think that solved the problem, that bouncing ifs or was it anything else? -- [[Axe]]<br />
<br />
It was solved with 1.9.4.3+. Yes, the wall-bouncing-when-close really helps. But I also shortened the blind-mans-stick at close ranges, which makes P go less steep at the enemy when in close combat. It also gives me a wall-avoidance problem though which I am not sure how to counter. I think that on average P hits the wall two times each round. Not good!! -- [[PEZ]]<br />
<br />
Truly strange that i 1.9.7.6 doesn't perform like 1.9.7. Obviously I don't understand my own code... -- [[PEZ]]<br />
<br />
I still don't get it. In 1.9.7 I fire an EnemyWave every tick. And if the enemy has fired I flag the wave as surfable. Then I only use surfable waves to update my reverse and forward danger guesses. In 1.9.7.6 I instead only fire enemy waves when the enemy fires. No need for a "surfable" flag then. This is the ''only'' difference. Apart from almost 20 rating points that is... -- [[PEZ]]<br />
<br />
Ha! Another 2K member, surelly feels good, ain't? Congratulations! Register it in [[The2000Club]]. -- [[Axe]]<br />
<br />
I just did. And it sure feels good. Do I need to remind you P is a mini? =) -- [[PEZ]]<br />
* I think u just did it :) -- [[Axe]]<br />
<br />
(edit conflict!) And is also the first member of [[The2000Club/Mini]]! What was afterall the feature that gave u the 2K? -- [[Axe]]<br />
<br />
"The feature" was to go back to [[Aristocles]] gun. It was that strange targeting bug that gave me the hint. It seems that the wall segmentation is really important. I'm tweaking it as of now and I think I can improve the gun a little more. We'll soon see if I'm right. [[Apollon]] was the first member of that club though. Depending on the entry rules I guess. -- [[PEZ]]<br />
* Why? Did Apollon reached 2K with 500, but dragged down after or it reached only in the mini score tables?<br />
* It has 2K+ points in the minirumble. I think that's the entry ticket for that club.<br />
<br />
You did it! Great stuff, congratulations! :) -- [[ABC]]<br />
<br />
Thanks! Glad I took that screen shot while I was still above 2K. It seems that fighting minis only drags my rating down. I just might release my first MegaBot in months just because of this. =) -- [[PEZ]]<br />
* I really don´t understand it , please explain it slowly (i´m a surfer), by beeing a mini the client priorize the battles against minis? -- [[Axe]] <br />
<br />
Yes, the client makes sure every bot in every class gets 500 battles. -- [[PEZ]]<br />
<br />
Let me clarify that. First the client fills up a bots 500 battles in the general rumble. Then, if the bot is a mini, the client starts running mini battles until the bot has 500 battles fought there. (And this goes on if it's a micor or nano.) This means that after 500 rounds Pugilist had the general mix of battles. But then battles against minis start to get over representated. I don't quite understand why this lowers the rating for P, but it seems to be quite a consistant pattern. -- [[PEZ]]<br />
<br />
Congratulations on joining [[The2000Club]], [[PEZ]]!! It is well deserved since you are probably the most persistent robocoder around. --[[Vic]]</div>Voidious