i'm always happy when I see entropy being used outside thermodynamics (I could never understand it in my physics / thermodynamics classes; finally picked it up when learning information theory!).
i was thinking of one more thing - cna we make this conditional on the ball (assuming you have distributions on lines / lengths)? or are you doing this already (for a full-pitch leg stump, 40% is sweep, but this batter only sweeps 20% and so high cross-entropy etc)?
Yes, this already does that. For each (line, length) bin, we construct distributions of shots and compute the self and cross entropy. The overall entropy and cross entropy values are the weighted means of these values over the particular batter's (line, length) distribution.
i'm always happy when I see entropy being used outside thermodynamics (I could never understand it in my physics / thermodynamics classes; finally picked it up when learning information theory!).
i was thinking of one more thing - cna we make this conditional on the ball (assuming you have distributions on lines / lengths)? or are you doing this already (for a full-pitch leg stump, 40% is sweep, but this batter only sweeps 20% and so high cross-entropy etc)?
Yes, this already does that. For each (line, length) bin, we construct distributions of shots and compute the self and cross entropy. The overall entropy and cross entropy values are the weighted means of these values over the particular batter's (line, length) distribution.
Also would be useful for bowlers and conditioning on them -- what balls do bowlers usually ball/how many do they concede, and how does that vary?