The Geopolygonizer takes about ~5 hours to run on the oregon landcover raster on a Linux machine with 12 processes. This is okay as long as we expect not to have to do this often or not have to run on much more data. Eventually, however, we may want to run on more datasets, across a wider region, more frequently.
When this happens, we should consider parallelizing the steps to run on a GPU. Fortunately, a lot of the steps involve a bunch of for-loops or numpy logic that I imagine with some care, can be managed to run correctly in this manner.
The Geopolygonizer takes about ~5 hours to run on the oregon landcover raster on a Linux machine with 12 processes. This is okay as long as we expect not to have to do this often or not have to run on much more data. Eventually, however, we may want to run on more datasets, across a wider region, more frequently.
When this happens, we should consider parallelizing the steps to run on a GPU. Fortunately, a lot of the steps involve a bunch of for-loops or numpy logic that I imagine with some care, can be managed to run correctly in this manner.