tysen2k

So VM emailed me to bring me into this discussion... Unfortunately I don't have much to contribute besides what I posted about 6 years ago. I haven't really looked at it since. Searching for my name and zar points should bring up a lot of stuff. And as I've said several times before, I think the interesting stuff on hand evaluation isn't on who has the most accurate system down to the last decimal point. What's interesting is quantifying how your evaluation changes as the bidding progresses (the auction in the OP is a perfect example). Other interesting stuff that I always wanted to look into are things like how high cards and distribution really aren't additive and how the relative importance between strength and distribution changes depending on how balanced partner is (and the opponents). I feel there's a lot of potential here for someone to research, but I just don't have the time. Tysen

That's an interesting idea. It seems like a lot of work and would be a good similarity measure, but it still doesn't tell you if one set of opening bids is "better" than another. Any ideas on that front? I'm still working on Silent Spade and it's making some good progress. Once I got it over the hump of never opening up with higher bids, it's now taking off on its own. It's slowly refining itself each day and hasn't peaked yet. Here is the current iteration (the spades aren't so silent anymore): Pass –0-9 HCP, 3-5 spades, 1-5 in other suits OR 10+HCP, 4-5 spades, 1-4 clubs, 1-5 in reds but no 4441 shapes. 1♣ – 0-12 balanced with 2 spades or 10+ balanced with 3 spades 1♦ –any 5-4 or 5-5 hand that isn't passed, or 10+ HCP any 4441 1♥ – any hand with 5+ hearts that can’t open 1♦/2♣ 1♠ – any hand with 6+ clubs that can't open 2N/3♣ 1N – any with 6+ spades, 0-4 clubs 2♣ – any with 6+ hearts, 4-5 clubs Higher bids are all distributional 1- or 2- suiters with no HCP dependence The computer doesn't like the 1♣ definition that much, so it seems that the room for improvement will come from there. Tysen

Dbl and 1♦ should always show solid values since they give the opps extra steps. I've used them to show solid heart and spade overcalls respectively. Tysen

The Silent Spade is still mostly silent. I haven't put too much time into it, but I'm trying some tweaks to try and make it use some of the higher opening bids. My current iteration now bids up through 2♣, but I'm slowing getting it to bid higher. One thing that I am seeing is that even though I always start off with random initial conditions, it almost always gravitates towards pass = spades, 1♣ = balanced. The current iteration is approximately: Pass - 0-37, all hands 4+ spades unless qualifies for 1♣, 1♦, or 2♣. Also pass with any 14+ balanced. 1♣ - 0-13 balanced with exactly 2 or 4 spades 1♦ - 0-37, all unbalanced hands with 5+ diamonds unless spades are equal/longer. Prefer 1NT over this if qualified. 1♥ - 0-13 balanced with exactly 3 spades. 1♠ - A strange one. Again about 0-13, 1-2 spades, 2-4 of both red suits, 5-6 clubs. 1NT - 0-37, 7+ clubs or 11 cards in the minors. 2♣ - 0-37, 6+ hearts. If has exactly 4 spades, bid 2♣ with 0-10HCP. Tysen

I have not had much time to investigate this problem further, but I'm still having issues with my model in the sense that I don't get "normal" preempts. It may be a combination of two things: There is a severe problem with local minima. Since the entropy function is minimized when similar hands are put together, it tends to group closer to its initial configuration and not want to move in a new direction. Trying to increase the opponent's entropy might not be the best function to determine the right preempts. The classic way to deal with the local minima is to start from several random starting points and see which leads to the best outcome. Unfortunately my current setup is way to slow for this. One solution to local minima may be to use Helene's method of decision trees. I think you'd need to divide it into at least 128 or 256 groups since preempts can be very rare/specific. The value function for the decision tree could be our entropy minus opp's entropy, although there may be problems with this (see below). Here's how I would do the hill climbing to take into account bidding space. Once you have the leaves of your tree, randomly pick 11 leaves and assign the opening bids Pass, 1C,..., 2N (that's 11 different bids) to them. Figure out the entropy to those 11 leaves. Leave pass alone, but give an entropy penalty of 'x' to 1C, '2x' to 1D, etc. Then do hill climbing allowing other leaves to attach themselves to one of these bids. Repeat with several different initial 11 leaves. We can vary the x penalty to make sure we get a distribution that we think is normal (pass should be 40-50% of all hands, 1C should be 20-25%, etc.). Also, feel free to pick a different number of leaves than 11 if you want. On to point #2 about my preempts. I tried several different starting points (including seeding it with "standard" preempts), but I found that it always wanted to do things like open at a high level with strong hands with long diamonds. The value function is finding hands where it can be relatively sure of its own contracts, but also eating up space when the opponents have a good contract too. And all that seems logical and the whole point of preempting. However, when we're looking at their "perfect contract" that contract is sometimes a sacrifice. Their perfect contract of 4S might be because they can make, or because they have a good sacrifice. So my model wants to open high with specific good hands like: - Ax AKJxxxx AJxx because it has the logic of "my opponents can probably preempt against me, so I better describe my hand really well now before they get a chance." Kind of preempting a preempt, but with a strong hand. The model seems to think that using the high bids for these kinds of hands is valuable and puts the normal preempt hands into the constructive bids. Maybe it's on to something, but I'm thinking probably not. That's it for now. Tysen

Okay, here's an interesting thing that I noticed about this. Look through the tree at all the places where it asks for information about the same feature twice (It asks about spades when it already asked about spades previously). I count 32 times where this happens. In 23 of those cases (72%) the question was asked again when a low answer was given to the first question. So "I'm weak" "Really? How weak?" was more common than "I'm strong" "Really? How strong?" Same with "I have short spades" "Really? How short?" Is this an artifact of the setup, or is there more value in telling partner what you don't have? Tysen

Let me just say that I'm glad we're brainstorming here and all coming up with different ways of handling the same problem. I'd like to encourage others to try their own methods too. Tysen

There are plenty of dummy variables that you can add besides raw suit lengths. Length of longest suit Length of longest major Length of longest red suit Length of shortest major Number of doubletons Number of singletons Number of voids etc. Tysen

Do you propose solving auction termination in general? Only in a relay context? Are you talking about something that can handle any starting level and any known info about partner's hand? Tysen

Yes, that's what I'm doing now by looking at the chance we can make ours and the chance they can make theirs. I haven't had any time to work on it this weekend though. Tysen

I'm running off of 20k training cases. I've got 20 nodes that are being fed into a softmax function to determine the opening bid. I've tried increasing the number of nodes and it doesn't seem to make much difference. I've recently started trying to fully incorporate competition/preemptive bidding. I'm now also looking into what the opponent's best contract is in addition to our own. So I'm now trying to maximize the chance that we find our best contract minus the chance they find theirs. I let it run overnight... And I'm still getting results similar to when I had no competition. It never wants to bid higher than 1N and the bid definitions are somewhat similar to the original Silent Spade, though they use more HCP info. Pass is something like 4+ spades (unbalanced only) or 15+ HCP and both minors 1C is 0-12 HCP, no singletons/voids, 2-4 spades, 2-5 hearts 1D is clubs 1H is diamonds 1S is hearts 1N is 0-10 HCP, semi balanced, usually has diamonds and only 2-3 in both majors It's really strange... I'm going to look at it some more and see if I can find out why it's not preempting at all (except that 1N bid is a preempt). And I don't know why it likes "1 over" transfers either. :rolleyes: Tysen

Right, which is actually good because it's another component of competative bidding the model isn't ignoring. If it were to only focus on constructive bidding, it might want to stop in 2S when the opponents can make 3H. Tysen

I too am still skeptical about the process, but I thought it was cool enough to post anyway. BTW, can everyone still see the graphic that describes the system? I could see it when I posted it at home, but now that I'm at work it doesn't show up. We do have lots of filters here, so it could easily be blocked for just me. [Edit: Helene, thanks for the confirmation it still shows!] The HCP indicated are just sampled exact figures, so if there's a transition between two HCP squares, it happens somewhere in between those amounts. The posted system is for 10% efficiency. Nick, I totally see your point about efficiency, but I'm not sure if there's a fair way to handle it. Let me think about it some more. Actually, maybe trying to generate responses to these openers might be a good validation. I'm worried that there are too many hands that pass and bid 1C and it may not be able to come up with good responses. I could try that next. Tysen

Here is the system in all its glory: http://img819.imageshack.us/img819/3413/bridgenn.jpg Okay, we’re looking at 4 dimensions here: spades, hearts, diamonds, and HCP. Clubs can be figured out once you’ve defined the other 3 suits. There are several 7x7 blocks of bids that each contain the same number of hearts and diamonds. For example, the 7x7 block in the bottom left are all the hands with 0 hearts and 6 diamonds. Within each block are all the subdivisions by number of spades and HCP. The colors correspond to the opening bid spelled out in the legend. I figured this is a nice and compact (and colorful!) way to show all the possible hands. Tysen

Actually I started off with a 20% efficiency. It had the opposite effect of what you are hoping for. It wanted to keep its opening bids even lower because it thought that the responses would be much better. It didn't want to eat up any bidding space at all and reluctantly bid up through 1S. Tysen