[Gilithin]

mycroft, on 2023-May-12, 08:46, said:

the judgement required when you know "3361, any 9+, if I ask any more we've lost 3NT"

If you reach that point and do not know whether you want to be in 3NT, you probably should have made a relay break to find out on the previous round. If the relay system does not allow such breaks, get a better system.

[mycroft]

3NT or 6♦, of course, not "is 3NT going down?" I'm guessing that partner with enough for 7 will not pass 3NT :-)

The relay system I'm familiar with asks about specific cards using Spiral Scan (okay, some relay breaks to look for one specific card, but still) and even if I don't care whether partner is 3361, 3163, or 1363, Spiral does kind of rely on partner knowing exact shape...

Of course, if all you ever play is high-level IMP matches, "having to" play 5♦ for lose 1 or 2 isn't a big deal when partner does have the wrong/one too few cards. That's a decision point, too. OTOH, there was that time I played the 2-day (okay, for me, 1-day) National Open BAM...

But we really should be asking Hrothgar about this, given the givens.

[pescetom]

mycroft, on 2023-May-12, 08:46, said:

Add to that that there are just people for whom lots of memorization, especially with sensible guiding principles that can be used to regenerate the path at the table if necessary, is low-energy. I am one of them, at least I was 15 years ago.

I quote that, of course, but one of my regrets is that I did not play bridge 15 years ago.
At least I am still prepared to give "lots of memorization" a go, unlike current potential partners, many younger or apparently sane.

[awm]

Maybe a stupid question, but where do you get the hands you’re clustering? My concern is that if you are creating random deals, it’s possible in principle to place the contract looking at only one hand since you thus know what deal it is and thus the other hands. It might work better to generate a set of 100 “player A” hands accompanied by 100 “player B” hands for each rather than just 10k random hand pairs.

[helene_t]

awm, on 2023-May-13, 15:44, said:

Maybe a stupid question, but where do you get the hands you’re clustering? My concern is that if you are creating random deals, it’s possible in principle to place the contract looking at only one hand since you thus know what deal it is and thus the other hands. It might work better to generate a set of 100 “player A” hands accompanied by 100 “player B” hands for each rather than just 10k random hand pairs.

Yes I had this concern also.
Here is what I did:
- deal 100 mio deals
- select those 38 mio deals where it's plausible that NS don't bid
- estimate the number of tricks in each denomination from the number of trumps and hcps. I could easily factor in shortness also, long suit bonus for notrump etc but texture info would give me too many different hand types
- reduce ad hoc to 1860 hand types for each of the two players, giving some 3 mio deal types
- for each deal type, calculate average tricks, and note the number of deals as that is going to be used as weight
- for deal types that could exist but had a frq of 0, I based the number of tricks on interval midpoint for suit lengths and hcps, and assigned a frq of 0 001. This is because I wanted those artificial data to be influential only for distances that couldn't be supported by any real data.