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

BGonline.org Forums

Cubeful equities: how to derive from cubeless?

Posted By: Tom Keith
Date: Sunday, 10 January 2010, at 7:15 p.m.

In Response To: Cubeful equities: how to derive from cubeless? (Philippe Michel)

Do you have some idea on how feasible it would be to create software that, given a database of positions, could extract the N most close (backgammon-wise) to some other position given to it?

The problem, of course, is how do you define "close"? I think I know what you mean. You're looking for positions that have a similar theme to them, perhaps so you can study them together. But how can you explain this to a bot?

Here's one idea I had that might be a step in the right direction. Backgammon neural nets are typically set up in three layers: an input layer (raw data from the board), a middle layer, and an output layer (final equities). Who knows what values are represented by the middle layer, but presumably they are some kind of higher level concepts that the bot, through training, has decided are important.

If you looked for positions that have a high correlation to each other of values of the middle layer of neurons, you might get positions that are similar in the way that you mean. Or maybe not. I haven't tried it.

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.