Test Results

Jump to navigation Jump to search

Test Results

The gun results are looking very strong, congratulations on the great results.

Time to be a downer however. These great scores do not necessarily result in improved rumble performance. Aside from the small sample sizes and the age of these challenges, it's mostly that the movement is far more important that the gun in terms of score. For example my robot Nene originally broke into the top 10 using Raiko's (the MC2k7 reference) gun.

Chase08:42, 26 August 2017

Agree. To improve percentage score, it is much easier to prevent your opponents from scoring than earning more points by yourself.

However, the gun is much easier to tune than movement, as there are fewer edge cases in gun than movement. And targeting challenges do help a lot. But the movement challenges are not so helpful.

Xor (talk)08:56, 26 August 2017
 

Yeah, thank you for your opinion!

I actually devoted too much time on my gun exactly because it's way easier to test. Every time I touched my movement code I went back to my gun 5 minutes later, i'm creating a lot of resistance to work on it. Anyway, I'm was planning to improve my movement based on the wiki challenges, but since you said that i'll probably look into alternatives. Notice that I do my work based on challenges not because I think these results are strongly related to rumble results, but because those are my first good results in robocode so I feel like improving over challenges is practical and is still improving my rumble results. Its clear that this is not true for my gun anymore.

Going back to the alternatives, do you have any suggestion? Should I start to build my own rumble-directed testbed besides the bots I perform bad after each release? There are bots which are like "a must"?

Rsalesc (talk)02:02, 27 August 2017

Just looking at your results and the differences against my bots, it seems that your distancing and avoiding the simple guns is not near perfect yet. F.e. against wiki.nano.DevelFISH and bbo.RamboT, both close fighters, your score is lower than expected. I usually only tested against an older version of my own bot, until I finally made a testbed with a few top-50 bots, white-whales and also some 150-400 ranked bots.

GrubbmGait (talk)10:56, 27 August 2017
 

Let me suggest the config from my Roborunner (posted below). While it does not predict the final APS, it very well show if APS would improve or drop compared to the base version. Several years of testing show that every time I was tempted to go with my gut feeling and against Roborunner prediction, I was wrong. So now I trust it.

Your bot seems to be in similar ranking to mine EvBotNG (except recent flop). So I expect it will have same prediction power.

Sample 1v1 Test Bed
PERCENT_SCORE
35

Hard Bots {
jk.mega.DrussGT 3.1.0
voidious.Diamond 1.8.22
lxx.Tomcat 3.68
}

Dream to win {
abc.Shadow 3.83c
}

Circular Targeting bot {
TCMI.Enyo 0.4
}

Head On bot {
rz.HawkOnFire 0.1
}

My League above {
mladjo.Startko 1.0
trab.nano.AinippeNano 1.3
ds.Versatile RB1.0.1
}

My League below {
#non.mega.NoName 0.0
# Czolgzilla takes to long to execute a battle
#deith.Czolgzilla 0.11  
}

Fighting close Bots {
nz.jdc.nano.NeophyteSRAL 1.3
fcr.First 1.0
}

Rammer Bots {
zyx.nano.RedBull 1.0
radnor.RamRod 1.0
}


#Easy Bots {
#sample.SpinBot 1.0
#sample.Crazy 1.0
#}
Beaming (talk)17:02, 27 August 2017

Thank you, gonna try that soon :P

Rsalesc (talk)02:58, 29 August 2017