To do this, I implemented the Heckman correction method using a custom likelihood in PyMC. Details here!

github.com/justinochoi/...

how about "projection onto the vector space spanned by the columns of the data"

This is arguably somewhat a team effect, since different teams will have different reports on what pitches to throw to whom (which the catchers use)

Would be interesting to see if this effect persists even if a catchers switches teams

running mcmc satisfies my dopamine cravings

I don't think it will make *that* much of a difference because the normal approximation to everything is good enough, but this is definitely more flexible and rigorous!

Happy birthday Ben!

Are the expected movement distributions a mixture of gaussians, one for each pitch type?