Here is what I am trying to do.
I have a set of points with coordinates. Each point has data associated with it, so each point can be shown for instance in a pandas as a row, with many features, including a pair of X/Y coordinates. There is a feature which will be the target, labelled 1 or 0. The goal is to be able to add a new point (anywhere within a region), and then get a prediction of 1 or 0 for that feature.
One difficulty is that the points are not evenly distributed. Some areas have many points, and some others not much. I think I remember @jeremy mentioning in the class somewhere that actually CNN are ok with sparse data. Am I remembering correctly?
I think I might use fastai tabular for that, but I wanted to know if somebody had done a similar approach with geographical points and maybe has some code to share.
Thanks in advance,