I can’t say I understand their example… I see they have a paper that is somewhat related, so I’d like to look into it further there. The values in x_data look like counts to me, but I don’t see where the location information comes in. The GP code in PyMC3 would expect something like:
X: 2D coordinates of the shot, eg. a stack of (x,y) pairs.
y: #of shots made at each coordinate.
So if X had dimension (100,2), y would have dimension (100, ). You could additionally index on player if you had multiple players’ shots.