General > Bot Challenges

combat challenge

<< < (3/4) > >>

Moonfisher:
Heh unfortunately Excalibur no longer works that well in the newest drops.
It still starts off ok, so it does provide a challenge, but it falls appart over time.
The newest lovebot can beat 50 of them if it gets realy lucky  (Had to remove a condition for the conspec though : *.totalbots 21 <)
But I think maybe Excalibur is too hard for this type of challenge. We probably need a bot which is custom made for this kind of challenge.
For instance 1 Spinner 1.1 kan beat 1000 reapers... (Ran it on a laptop, took hours, but it won). This is just because of one or two of it's genes that work realy well against a bot like reaper.
So it should probably be a simple bot, but it would need to have good defences against ties, viruses and venom, to make sure you can't exploit a weakness.
In the end it's a strange challenge, since you're just working on finetuning a bot to beat one specifik bot while outnumbered.
It's a cool concept, it's just hard to pick one bot to fight against. Maybe it shouldn't just be one oponent, maybe there should be 5 different bots with very different behaviors, so your total score is the combined amount you can beat in each category. Maybe you should beat atleast each category in a 1 vs 5 or something.
Or maybe you should fight 5 different bots with the same conspec, at the same time.
Still ends up being an odd challenge, but might be fun if the 5 bots to beat represent different challenges.

I think I want to look at that neural network challenge, saw the topic a long time ago but noticed someone had already done it... but I want to take a look at that bot, see what you could do with that...
I'm thinking if it's cool, then once I have more time I'll explore the idea of using a small C program to train a good strong bot into a neural network just using backpropagation and with enough neurons and training time to make it fit perfectly with the base bot.
Then just write all the weights into the bots code, as a fixed bot with fixed weights. And then only use point mutations or whatever it's called, to tweek the weights in the bot. I think it could be interesting if a complex top league bot could actualy be improved through evolution this way. Would be hard to set up the evo sim for it, since you need to evolve it against all the other bots in the league... would be cool if you could make a sawn gate and set up some timers and conditions for spawning a random oponent from a list, would also help prevent canibotism.

Moonfisher:
Heh Spinner v1.3 can beat 5 Multiply4  (It wins about half the time)
And it can beat 50 of them in a size 8 field with 80 veggies...
Was running DB 2.43.1h, since I and j have a bug causing most bots to make too much venom/poison...
Anyway I like Multiply4, it's a very strong F2 bot, so I often use it for testing, which means the bot I'm making often ends up doing well against it.
And they're in the same league (F2), although the new Spinner uses ties to kill and confuse... (No feeding ofcourse)
It can also beat 17 Excalibur some of the time... if it starts near an alge...
But it's hard to tell if it's valid results... since DB 2.43.1h has the bucket bug

gymsum:
Use the sea snake.

Moonfisher:

--- Quote from: gymsum ---Use the sea snake.
--- End quote ---

I will once version K comes out, J has the poison/venom bug and it's very agressive in that version. and Seasnake only works in version J (Since it was developped partialy do debug the ties, so some of it's features don't work in earlyer versions).

Also if we're going to keep using strong bots for this, maybe we should also add the option to see how many of your bot it takes to beat the target bot

EricL:

--- Quote from: Moonfisher ---I will once version K comes out,
--- End quote ---

Will be today or tomorrow.  It's mostly ready except I'm working on removing the max bots limit of 10000.  Doing this right entails changing the way the rob array works to make it size dynamically.  Should result in a smaller working set than today most of the time and a bot limit of 32000.

Navigation

[0] Message Index

[#] Next page

[*] Previous page