[Computer-go] Kassandra talking

2014-12-22 Thread Ingo Althöfer 
Hi all,

it seems computer-go faces exciting times again.
What a wonderful world.


Switch to Kassandra mode: Several years ago (ca 2008) 
Sylvain Gelly had his Ph.D. thesis and in it a section
on the quality of random game generators. One of his 
(experimental) findings was that better generators do
not necessarily lead to better play of the corresponding
MCTS engine.
Inspired by this I did some theoretical research (in 2009) 
for pure Monte Carlo (without tree part): Without problems
I was able to generate situations where better move generation
led to worse performance.

In total: Changing the random move generator typically will
change the playing behaviour. However, it can not be well
predicted if this change will be to the better or to the
worse. 
**

Only tests in practise will show. 

Ingo.
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Kassandra talking

2014-12-22 Thread Robert Jasiek 

n 22.12.2014 09:46, Ingo Althöfer wrote:

In total: Changing the random move generator typically will
change the playing behaviour. However, it can not be well
predicted if this change will be to the better or to the
worse.


Is this prediction theoretically impossible (why, under exactly which 
presuppositions) or is research in understanding it not advanced far enough?


--
robert jasiek
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Kassandra talking

2014-12-22 Thread Ingo Althöfer 
Hallo Robert,

  In total: Changing the random move generator typically will
  change the playing behaviour. However, it can not be well
  predicted if this change will be to the better or to the
  worse.
 
 Is this prediction theoretically impossible (why, under exactly which 
 presuppositions) or is research in understanding it not advanced far enough?

at the moment my impression is that simply our understanding of
MCTS is not advanced enough.

But your question is really interesting: maybe, someone can prove
an intractability result for the topic.

I will think about it.

Cheers, Ingo.
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go