draft roadmap

[mathause]

We (I) should draft a high level roadmap and possibly a time line...

@MESMER-group/mesmer

[mathause]

This is not yet the roadmap but I want to list some of the important points:

#30

• define high level internal data structure (now: nested dicts; future: ?) internal data structure #106
• low level data structures (now: numpy; future: xarray) unstructured grid #105
• save output in a well defined format (now: pickle; future: netCDF + yaml): Introduce ability to save output (i.e., calibrated parameters and emulations) to netcdf files #65, Change to saving bundle using more standard format #18
• define config infrastructure (now: module, no defaults; future: dict?) config state & default config #112
• disentangle the data wrangling from the statistics e.g. Add prototype of class structure #109

[mathause]

WIP:

Mesmer has two distinct phases

1. calibration
2. emulation

#106 (comment)

1. Start with data to which we want to calibrate (e.g. CMIP output, or some impact model output)
2. Find the data we want to use as inputs (e.g. CMIP global-means having adjusted for volcanoes first, extreme rainfall data)
3. Calibrate a model (e.g. regression, some other fancy stats stuff I have no idea about)
4. Store the calibrated model parameters (how to do the storage well is a separate point)
5. Use the calibrated model parameters to draw emulations