[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

The MCG challenge.

Posted By: Matt Cohn-Geier
Date: Friday, 4 December 2009, at 6:50 a.m.

In Response To: The MCG challenge. (Matt Ryder)

• Is this just a money-game proposal? Or are you prepared to play a series of matches against the bots? (Match-play would ensure that more games reach a point where the bot can use its bear-off database – not to mention its MET)

Money games are more favorable for humans, for reasons unrelated to the bearoff database and MET. But matches are also OK if they are long enough. But definitely prefer money games.

• Would you be prepared to play a series of 1-pointers, or do you need to lure the bot into high-value cube blunders to beat it?

The high-value cube blunders thing.

• How long would the participants have to consider each play? Do you claim to beat the bots if they’re allowed the luxury of a rollout, or is it simply their n-ply lookahead you claim to conquer? What is the search-space? What are the settings? What’s the maximum spec of the hardware your opponents can employ to run the bot?

I don't know. ROs may or may not be beatable depending on the settings. Hardware will only make it play faster rather than better since the strength has to be predetermined, i.e., the bot can't make a play after X seconds.

• The winner would be declared based on the best of a series of 31 5-point matches played over several days.

5-point matches are too short. Best-of-31 5 point matches is no good but best-of-5 31-point matches might be OK.

That sounds like quite a lot of effort to prove a point. Why not rather simply post a position where XG would erroneously skyrocket the cube?

Too cheap to buy XG.

More seriously, I might buy XG, although first I want to see what is going on with the licensing thing to X # of computers and XG 2.0+ and so on.

Anyway, I never intended to prove a point...as far as I'm concerned it's already been proven by Daniel & others. And I have seen XG's evals of outside primes & huge back games and they are not good.

To be convincing, it has to be a "real" position from an actual game, not a 'textbook' example artificially manufactured (such positions are often almost impossible to engineer in reality.)

You might be surprised at how easy some of these are to engineer (or, alternatively, how bad the bot's checker play is).

Messages In This Thread

 

Post Response

Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:

 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.