Guys - I'm now in branch mike_review2 in my fork --- sorry, long sad tail.
Jotting down ideas for future stuff here, mostly for part2:
Prior Monte Carlo - I like the idea of using the prior MC results either to trim sampling range based on quantiles (e.g. Neversink) or completely removing rejected realizations and doing a hotstart. Perhaps an alternative notebook between MC and iES
Also, thinking a little more discussion of the importance of the prior for iES in particular
Spectral Simulation - more about this would be rad
obs and weights - show differential weighting of PHI components and the use of a separate variance for noise obs
localization - should we incorporate some of JP's findings about the tradeoffs between sample size and localization? D'OH! I see this is mentioned already. Missed on the first round
Prior MC and localization notebooks - it seems like the prior with and without FOSM looks really similar. Not a huge advantage to the FOSM sampling. Similarly, the parameter changes prior to localization look very minimal. Wondering if it's because I ran the larger ensemble in the part 1 ies nb?
Should we incorporate a notebook on cloud computing? HTCondor? etc. for industrial strength runs?
For DA - we should probably make a version with the MF6 API....
Guys - I'm now in branch
mike_review2
in my fork --- sorry, long sad tail.Jotting down ideas for future stuff here, mostly for part2: