Closed basaks closed 5 years ago
May be something like this? https://rasterio.readthedocs.io/en/stable/topics/concurrency.html Can we do an mpi equivalent of the above?
Turns out this was introduced due to compression and tiling during file write. New code changes allow optional compression and file write via config file.
Some prediction benchmarks using
multirandomforest
and only 5 covariates on a national dataset for one partition (out of 100):Writing to disc take about 68% of the time.
The writing to the disc currently in uncoverml uses only one cpu. It can be improved using multiple cpus.
Here is one partition during the prediction: