Jump to content

Event 15 information and score reporting


smerriman

Recommended Posts

You have my full respect and appreciation for the work you do. However, FWIW if I were the TD I'd take the consequence of it and have a policy to amend the results of those few horrendous outliers - not to be misunderstood, this of course applies to any player in tournaments where the blunder counts.

We've tried to fix some of the worst bugs -- it used to double 6NT with 2 cashing aces, then not lead them! (Presumably it envisioned a small chance to get a 2nd undertrick if it didn't set up declarer's kings, and not enough of the simulated hands had 12 tricks available in the side suits.)

 

This particular problem is harder. It's not so obvious when the results of the simulations need to be ignored. And we can't look at the result at the other table to ensure that it plays the same way, to prevent

 

And if you think the situation you raised is bad, how about all the times that it leads its singleton trump king, which it probably would have won when declarer takes a finesse? This is a consequence of the use of double dummy simulation -- it "knows" that declarer will play for the drop instead of finessing. I've considered whether we should put in another special case to block this, but there's a reason why it thinks a trump lead is best, and in many cases it doesn't actually give away a trick (if the ace is over the king you were going to pick it up anyway, and if partner has the queen there's no finesse.), and we have to be careful when overruling these decisions.

 

Everyone who plays robot tourneys learns that things like this happen from time to time; the underlying design that uses DD simulation and statistical analysis has some obvious flaws. Sometimes you're the beneficiary, sometimes you're the victim. If things like this really bother you, you should stay away from GIB.

 

Our work on improving GIB is primarily focused on fixing problems in the bidding database. We actively welcome reports on these issues.

Link to comment
Share on other sites

We've tried to fix some of the worst bugs -- it used to double 6NT with 2 cashing aces, then not lead them! (Presumably it envisioned a small chance to get a 2nd undertrick if it didn't set up declarer's kings, and not enough of the simulated hands had 12 tricks available in the side suits.)

 

This particular problem is harder. It's not so obvious when the results of the simulations need to be ignored. And we can't look at the result at the other table to ensure that it plays the same way, to prevent

 

And if you think the situation you raised is bad, how about all the times that it leads its singleton trump king, which it probably would have won when declarer takes a finesse? This is a consequence of the use of double dummy simulation -- it "knows" that declarer will play for the drop instead of finessing. I've considered whether we should put in another special case to block this, but there's a reason why it thinks a trump lead is best, and in many cases it doesn't actually give away a trick (if the ace is over the king you were going to pick it up anyway, and if partner has the queen there's no finesse.), and we have to be careful when overruling these decisions.

 

Everyone who plays robot tourneys learns that things like this happen from time to time; the underlying design that uses DD simulation and statistical analysis has some obvious flaws. Sometimes you're the beneficiary, sometimes you're the victim. If things like this really bother you, you should stay away from GIB.

 

Our work on improving GIB is primarily focused on fixing problems in the bidding database. We actively welcome reports on these issues.

 

One of the problem I see with the bidding (at least for me) is the use of law of total tricks when one opens a weak 2 in fourth seat. GIB will raise you even though it may have a very weak hand and opp did not bid.

Link to comment
Share on other sites

Guest
This topic is now closed to further replies.
×
×
  • Create New...