has anyone implemented Algorithm 1 and Algorithm 2 from the paper?
Section 2.2 gives a very clear to understand poker example. I thought to bucket flop hands, i make histograms of them on the turn. Eg.: TcQd-7h9hQh would have a histogram
Such histogram could be represented in an array[50]. Something like {0,0, ... , 0.04, 0, 0.04, ...}
Then k++ mean cluster this flop histograms with a EMD distance function.
But they do something else what i dont really get.
Quote:
We note that Figures 9 and 10 depict the distributions of expected turn hand strength, as opposed to full distributions over distributions of river hand strength (since each turn card will lead to a distribution over strength in the next round, not a single value). For example, for the hand TcQd-7h9hQh, if the turn card is Ad, the hand’s expected equity, assuming a uniform random river card and uniform random hand for the opponent, is 0.695 (though it will vary for individual river cards); so the interval for 0.68–0.7 in the histogram would be incremented by one for that turn card. This is significantly more simplistic than our potential-aware approach, which takes into account the full distribution of turn ‘buckets’, which are themselves distributions over equity intervals after the river.
