I've been trying to use convex hulls to explore ML embedding spaces. However, the dimensionality (768+ dimensions) seems to crash established options like QHull[0], even with 64GB RAM (and 16 CPU cores, albeit libqhull is not multi-threaded).
Are there more appropriate algorithms for finding convex hulls where dimensions are ~768? Or any parallelized / GPU-optimized options that I should look into?
Are there more appropriate algorithms for finding convex hulls where dimensions are ~768? Or any parallelized / GPU-optimized options that I should look into?
0: http://www.qhull.org