-
-
Save rabernat/fd2522f1937a43f49ac0423d35915648 to your computer and use it in GitHub Desktop.
A naïve question : what is the 'face' dimension ?
The 'face' dimension is related to how the LLC models grids the sphere. https://xmitgcm.readthedocs.io/en/latest/performance.html#llc-reading-strategies
Is there a particular reason for using xrft and not an apply_ufunc approach ? The latter choice would provide a template for users to adapt the code depending on the analysis they want to perform.
No not really. We wrote xrft so we like to use it, that's all.
What do you mean by "full dataset is too big" when motivating subsetting? What happen if you decrease it?
The dask graph becomes too big, or the tasks require too much memory.
Do you have a description of the dask server (memory ...)? This would be useful for comparison with hpc.
For this I used 100 nodes, each with 2 cores and 22 GB of memory.
Is there a particular reason for using xrft and not an apply_ufunc approach ? The latter choice would provide a template for users to adapt the code depending on the analysis they want to perform.
The use of stack and group_bybins is just brilliant ! Thanks for showing this.
For your information, Shane used 80d long time windows in his paper.
What do you mean by "full dataset is too big" when motivating subsetting? What happen if you decrease it?
Do you have a description of the dask server (memory ...)? This would be useful for comparison with hpc.