an idea for endplay

[seebrand]

in endplay, Ben cannot find strip throw-in play, for dds always correctly determine the position of Q ,so all candidate cards have the same making probilty 1. but after following by next opps normal card(x), the candidate cards's making probilty of dummy or dealer become 0.5 /0.6 or 1 ( correct other suit) in the same round. i have an idea:to superimpose the probabilities of making a contract with the cards in hand and the cards in the dummy, so as to optimize the selection of cards. This method may result in slower card play, but it is more accurate， solving the difficult problem of playing cards with dds, such as discovering Queen..

[ThorvaldAagaard]

You are very welcome to try to implement it. Let me know if I can do anything to help.

If the idea is right, I think a lot will be interested, especially https://github.com/Ark223/ACE

[seebrand]

Perhaps it can be simulated in the current card playing process? Add a process: Assuming the next player plays a small card, calculate the probability of dummy's card making contract, add it to the probability of each card in self hand, and select the card with the highest total probability. In CardPlayer, what's the impact of PIMC? does it affect the score of each card? how about simulate and calcaulatie next hand's probility?

[ThorvaldAagaard]

First problem is to identify this specific situation.

Second problem is introducing state, as all play is stateless.

I have been looking at Bayes Theorem to add probability to each possible layout with 4-5 cards left.

So independent of BEN you can create a solver, that can find best play with 3 cards left in each hand and with a complete play.

I have also been looking at alphamju implementation for endplay, and it should be possible.

If and when it is solved I am sure I can find a way to integrate it into BEN.