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

BGonline.org Forums

Bot Comparison redux. Rollateers needed.

Posted By: Michael Depreli
Date: Sunday, 14 August 2011, at 4:28 p.m.

In Response To: Bot Comparison redux. Rollateers needed. (neilkaz)

Is the same set of 500 games being used or a new set?

Same set

As for setting the max to 46656 games, I think CI will almost always converge to less than .005 sooner than that even in complicated backgames, but it is a good fail safe stopping point.

Often one candidate play will reach .005 CI considerably sooner than another candidate play. May I presume that RO's for that candidate stop while the bot continues the RO for the other candidates until they reach .005?

Xavier is the best person to answer this.

What about race DB? Different rollateers may have different settings. I assume just use it if you have it since it enhances slightly.

Xavier has produced a document showing exactly how to set up the rollouts so that everyone uses the same settings.

During my RO's for the last test, I noted a few times that immediate GNU 2 ply cube errors could mess up the results. (I'd tell Michael if I noted this) For example, the bots chose different plays, and all plays should go D/P but GNU was taking for once choice, thinking it was a .98 take when the RO result said 1.04 pass. As a result, that play may have been scored as .04 worse than the others when in reality all the plays were the same emg. Hopefully using a stronger bot and a stronger cube setting (XGR) will minimize the chances of this happening. Anyhow, it would really be overkill to RO everything as double/no double, I think.

Whilst preparing to redo the rollouts myself, Frank, and Xavier isolated the suspect rollouts and reran them using XG and updated the final table.

Rather than publish that table it's been decided to enhance the whole project with stronger rollouts.

Once done they will be rescanned for suspect rollouts but hopefully there won't be too many but as you said where cube errors may influence

the results stronger and or double / no double rollouts would we worth doing.

Michael

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.