[DavidKok]

thepossum, on 2023-March-27, 00:09, said:

I thought I spotted a few minor errors in that dealer script but not sure.
I am trying to follow the modern style of 1NT

I'd love to hear them =D I'm slowly building up a base of scripts for common situations (open 1NT, they do/do not overcall at the 1 or 2-level, takeout doubles in various situations, preemptive bids). If there's a mistake I want to correct them early.

[thepossum]

Spoiler

You inspired me to try to replicate your Sim

I am sure I have errors but for Vulnerable I get - 5 points for blasting versus invite - using the parameters above
EDIT I did indeed have an error and my results are getting closer to yours - but there is a fair bit of variance with 1000 deals

I used bdeal with 10000 hands

Chance of error in mine is high since I manually coded up the scoring without using a score function

But I did try to faithfully reproduce your code and its the first such long script I have attempted

[thepossum]

DavidKok, on 2023-March-27, 02:17, said:

I'd love to hear them =D I'm slowly building up a base of scripts for common situations (open 1NT, they do/do not overcall at the 1 or 2-level, takeout doubles in various situations, preemptive bids). If there's a mistake I want to correct them early.

[spoiler]
I did say I wasn't sure. I think I misread something about the AceKing singleton - which ended up being ok

Now searhing for errors in my attempt to replicate

It was a very elegant and interesting script that made me think it was worth looking at and trying to replicate

Your name is in the comments at the top

Debugging my attempt is proving something of a challenge
[/spolier]

[thepossum]

Hi David

Did you make any exclusions for EW hands which would interfere and prevent extreme scores? - edit sorry

I am getting close to similar results now

[DavidKok]

I did not. In my previous runs (e.g. frequency analysis of certain Dutch Doubleton, Precision and IMPrecision auctions) I included interference clauses - i.e. gave a definition of certain second and fourth seat overcalls and excluded those deals. I also previously included second, third and fourth hand analysis by conditioning on relevant other hands not having a preemptive or natural opening bid. At some point I plan to include this in all my questions, but for now I've stuck with the easiest example.

Also some new fun simulation results. Enthusiasts of statistical analysis, Type 1 versus Type 2 errors and conditional expectations might recognise this. I've written some scripts that check the HCP of North+South, conditional on them being able to make a slam. The results are (un)surprisingly low.

• 6 of a trump suit makes (double dummy) but 6NT and higher fail. 2500 examples produced (69554 deals generated and solved double dummy, runtime 5049s). Average HCP: 25.86.
  
• 6NT makes but 7C and higher fail. 1500 examples produced (90207 deals generated and solved double dummy, runtime 6419s). Average HCP: 29.30.
  
• 7 of a trump suit makes (double dummy) but 7NT fails. 2500 examples produced (315361 deals generated and solved double dummy, runtime 22411s). Average HCP: 27.60.
  
• 7NT makes. 1500 examples produced (204954 deals generated and solved double dummy, runtime 15407s). Average HCP: 30.81

Of course the reason that the averages are so low is that we are so far in the tail of the combined HCP distribution that 'freak' slams making double dummy on low combined strength are actually more frequent than sound slams. For example, in the 2500 trump suit grand slams the highest combined HCP was 35 - apparently all the 36+ HCP deals that made 7X also made 7NT. I plan to do some further analysis on this (e.g. split the frequency of the HCP by the base rate of getting that HCP). I doubt this is useful, but it's fun and easy to program.

I've also ran a bunch of simulations on aggressive 2-level preempts, but I'll tinker with them some more before drawing significant conclusions from them. It seems to suggest that, when not vulnerable, opening frequently on 5-card suits is not aggressive enough.