I am requesting the addition of parallelization in the pyTEMlib.eds_tools.fit_model function for use on an HPC.
The residuals function may simply need to have spectrum input confirmed as a dask array and then either converted or add .compute() for dask to run it in parallel.
There may be other room for improvement including where some of the mathematic operations could be converted to dask versions such as da.dot, or da.abs, but I am not sure how much speed improvement this would add.
ahoust17
commented
3 weeks ago
I am requesting the addition of parallelization in the
pyTEMlib.eds_tools.fit_modelfunction for use on an HPC.The residuals function may simply need to have spectrum input confirmed as a dask array and then either converted or add
.compute()for dask to run it in parallel.There may be other room for improvement including where some of the mathematic operations could be converted to dask versions such as da.dot, or da.abs, but I am not sure how much speed improvement this would add.