| Poll |
| What is BBB? |
| Big Bad Boomerang |
|
4% |
[ 1 ] |
| Bubble Bath of Bermuda |
|
18% |
[ 4 ] |
| Best Bid for Boromir |
|
9% |
[ 2 ] |
| Purple Percimon Pie? |
|
68% |
[ 15 ] |
|
| Total Votes : 22 |
|
VoodooMike
Joined: Nov 07, 2010
|
 Posted:
Mar 12, 2012 - 14:28 |
 
|
| plasmoid wrote: | Looking one thread down I'm guessing Blood Bowl Bot.
Neural network or standard programmed AI? |
If you're talking about an AI to play the game, then man, only someone not familiar with neural networks would ever ask that.
| Hitonagashi wrote: | | You know you are a true geek when you spend a Friday night in Bugmans, drinking lots of beer, debating whether neural net or standard AI would produce the best results when applied to Bloodbowl. |
Yeah, good luck on training the neural network - you'd pretty much need a really good standard AI system in place to be able to train the neural network using the abstracted data system the standard AI used... even then, you probably wouldn't end up with something that could reliably beat your standard AI. |
|
|
Hitonagashi
Joined: Apr 09, 2006
|
| Posted:
Mar 12, 2012 - 14:41 |
|
That was my argument VM  . I forget all the details of Shriekbob's approach, but it was certainly interesting to hear how he'd approach it. |
_________________
http://www.calculateyour.tv - an easy way to work out specific team builds.
 |
|
Garion
Joined: Aug 19, 2009
|
| Posted:
Mar 12, 2012 - 14:42 |
|
@ Voodoomike - Grod has developed a bot to play against already. So it is always possible that this could be what BBB is. It may get used for the academy division so new coaches can learn to play the game against an opponent that is not too good at the game and so they can always get a game without having to wait about for another opponent in the A division. |
|
|
Woodstock
Joined: Dec 11, 2004
|
| Posted:
Mar 12, 2012 - 15:25 |
|
The new Academy has no bots, Christer is waiting for something else before the division becomes CRP. Not to mention that Grod is not working on a CRP bot afaik.
I believe BBB is the Box equivalent of Ranked Smacks. |
|
|
On1
Joined: Jul 12, 2004
|
| Posted:
Mar 12, 2012 - 15:59 |
|
Stop making me so damn curious... |
|
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Mar 12, 2012 - 16:03 |
|
| Hitonagashi wrote: | | That was my argument VM. I forget all the details of Shriekbob's approach, but it was certainly interesting to hear how he'd approach it. |
Regardless of how ridiculous it'd be to try to properly train a NN to play the game, it's not the type of AI that is best suited to this sort of game - you can't modify the proficiency that a neural network plays with unless you're using different snapshots of its learning progress. So, great if you're trying to win some competition and can think up an abstraction system to allow your NN to treat board configurations and turns as patterns... but for most games you want to have configurable difficulty levels for the AI.
| Garion wrote: | | @ Voodoomike - Grod has developed a bot to play against already. So it is always possible that this could be what BBB is. It may get used for the academy division so new coaches can learn to play the game against an opponent that is not too good at the game and so they can always get a game without having to wait about for another opponent in the A division. |
Grod's bot is LRB5. I'm not sure how useful it is to have Academy train people to play an older version of the game. |
|
|
Garion
Joined: Aug 19, 2009
|
| Posted:
Mar 12, 2012 - 16:05 |
|
yup it was built on an older client. But I bet it could be moved across without too much difficulty. However that has already been mooted anyway. |
|
|
Kryten
Joined: Sep 02, 2003
|
BALL BALL BALL! The first tournament where three balls are kicked off instead of the usual one, with the first goal scored ending the current drive. |
|
|
BooAhl
Joined: Sep 02, 2004
|
| Posted:
Mar 12, 2012 - 17:21 |
|
| JimmyFantastic wrote: | Black Box Bash.
A special tournament where there is no ball.
The first team to clear the pitch wins the match.
This would be AWESOME!!! |
What? Didn't that come last year? I thought the ball was some kind off bug. |
|
|
Christer
Joined: Aug 02, 2003
|
| VoodooMike wrote: | | Hitonagashi wrote: | | That was my argument VM. I forget all the details of Shriekbob's approach, but it was certainly interesting to hear how he'd approach it. |
Regardless of how ridiculous it'd be to try to properly train a NN to play the game, it's not the type of AI that is best suited to this sort of game - you can't modify the proficiency that a neural network plays with unless you're using different snapshots of its learning progress. So, great if you're trying to win some competition and can think up an abstraction system to allow your NN to treat board configurations and turns as patterns... but for most games you want to have configurable difficulty levels for the AI. |
If I were to implement a blood bowl AI, I would most likely use a more memetic approach, optimising for various goals ("get the ball", "make a TD", "bash the opponent", "stall") on a medium level. If anything, I'd use a NN for controlling when and how to switch between these types of behaviours (possibly using a genetic algorithm to teach it).
Difficulty levels could be implemented with a random factor added to probabilities for actions, and using different controller NNs (older generations if the genetic approach was taken) to have it do "less optimal" choices.
Mind you, AI is not my field and I haven't attempted this type of AI at all. The most complex AI I've made was an othello game which isn't exactly the same type of game as Blood Bowl. |
|
|
Hitonagashi
Joined: Apr 09, 2006
|
| Posted:
Mar 12, 2012 - 18:43 |
|
That's what Shriekbob was talking about .
I've written a Chess AI before, and that was fun...I would write a Bloodbowl AI, but my Java(for Grod's client) is extremely rusty, and writing my own client is time consuming. If Kal opened the connection to allow us to write an AI "Client" to the server, I'd probably have a play with that, and put some Ruby bindings in place. That however, has to be waaaaay down the priority list, as there's not many of us who would do that  .
My prefered approach would be a "probability cloud", optimising to control space on the pitch, so attempting to gain the most zones of control by walling opponents you could. You know the exact probability of a block removing a given player from a field, so you could spin that into the evaluation function for the position. That said, you'd also need some sense of tempo in it, and increasing the value of holding squares closer to their endzone relative to the time remaining in the half. I think you could make quite a good generic AI using those concepts.
Research into Go AI's I suspect would provide an interesting way of writing a Bloodbowl one, as I see them as very similar games. |
_________________
http://www.calculateyour.tv - an easy way to work out specific team builds.
|
|
VoodooMike
Joined: Nov 07, 2010
|
| Posted:
Mar 12, 2012 - 20:06 |
|
| christer wrote: | | If I were to implement a blood bowl AI, I would most likely use a more memetic approach, optimising for various goals ("get the ball", "make a TD", "bash the opponent", "stall") on a medium level. If anything, I'd use a NN for controlling when and how to switch between these types of behaviours (possibly using a genetic algorithm to teach it). |
A pointless exercise in complexity, using a neural network. 99% of neural network usage is people going "hey, know what'd be cool?" and then slapping them in where they don't really improve the solutions, just to say one is there.
Game AI for games that have any level of per-turn intricacy are not where they shine.
| christer wrote: | | Difficulty levels could be implemented with a random factor added to probabilities for actions, and using different controller NNs (older generations if the genetic approach was taken) to have it do "less optimal" choices. |
Which sounds like you'd be using a NN in place of a random number generator, more or less. Since the actual complexity of the choices would almost certainly be in the task-specific functions, there'd be little room for an NN to do much learning, or to apply itself to superior solutions.
| christer wrote: | | Mind you, AI is not my field and I haven't attempted this type of AI at all. The most complex AI I've made was an othello game which isn't exactly the same type of game as Blood Bowl. |
You might be surprised, really... sometimes it just comes down to puzzling out a good abstraction, and to let the computer think like a computer instead of trying to make it think like a person. |
|
|
pac
Joined: Oct 03, 2005
|
| Posted:
Mar 15, 2012 - 12:47 |
|
BBB means Black Bug Room. Sorry 'bout that. 
Edit: Just kidding. That memes something else. |
_________________
Join us in building Blood Bowl Sixth Edition.
In other news, the Hittites are back. Join us in #fumbbl.hi |
|
PurpleChest
Joined: Oct 25, 2003
|
It is the name of a tournament format coming soon to Blackbox. |
_________________
Barbarus hic ego sum, quia non intelligor illis -Ovid
I am a barbarian here because i am not understood by anyone |
|
|
|
| |
DIBBL Awards
Search
