Presently, the comparison spaces are constructed using the same number of points as the dataset.
If the diffusion time is reasonable, only a fraction of these points will be used -
so why not estimate the number of points needed ahead of time? It could save a significant amount of computation when powering the matrix.
Presently, the comparison spaces are constructed using the same number of points as the dataset. If the diffusion time is reasonable, only a fraction of these points will be used - so why not estimate the number of points needed ahead of time? It could save a significant amount of computation when powering the matrix.