Open JimCircadian opened 2 years ago
This ideally should be trialled with better benchmarks for performance from the existing pipeline runs, so consider implementing better analytics
There's not a lot of call to do this without performant infrastructure to support it, which we are still in the process of acquiring, so unassigned until we can look at it sensibly.
Data loading related:
https://docs.nvidia.com/deeplearning/dali/user-guide/docs/index.html
Leaving here visibility, not planning to use atm.
A batch generator for Tensorflow/PyTorch from xarray DataArrays/datasets:
https://github.com/xarray-contrib/xbatcher
and, a relevant discussion thread:
https://discourse.pangeo.io/t/favorite-way-to-go-from-netcdf-xarray-to-torch-tf-jax-et-al/2663/2
https://www.noahbrenowitz.com/post/loading_netcdfs/