jonking93 / dash Goto Github PK

You can recycle any kalman gain that uses the same "obs" array. Potentially large speed gains for successful PSMs with a joint update.

Any VS-Lite standardization will change as multiple proxies are updated. This is because the mean of each run can change as the analysis updates.

Perhaps there should be a baseline standardization given at the beginning of each PSM?
This would be along the lines of: Do a bulk Ye calculation before starting dash, then save the mean and std of this bulk calculation to use in all dash updates.

estat is super slow. Either speed it up, or have Npdft just accept iterations and then calculate once at the end.

1. Make a more intuitive directory format
2. More intelligent use of varDex, sequences, etc. Most of these functions should be called a single time and pass the outputs.

It made the joint ENSRF much cleaner. But might want to eventually add it back in.
The alternative is building a second PSM array.

Implement dash to handle ensembles that do not fit in working memory.

1. Should be able to couple multiple variables with a single function call.

2. Should have a couple 'all' flag

3. Perhaps automatically couple variables in a finalizeDesign function.
   (But this could be questionable, how to set state seq and mean indices?)
   Maybe have a copyIndices function for vars and template?

PSMs that are time dependent work fine for serial, but not joint.

PSMs that incorporate proxy memory are completely unsupported.