The 25-getting-started notebook should be a good starting point.

Add .gitignore file to ignore Jupyter Notebook checkpoints, pycache, figures, pkl files, and specific directories

To be honest, I have made a mess by using far too few git commits. Will
try to tidy things up before the next paper is submitted.

EFSOI stuff is now reaching usable status for the LETKF paper. Also
found that when mirroring the gaspari cohn I had one index wrong.

The weight matrix where no obs are present was set to zero instead of
to a diagonal.

as slow as it use to be.

Adding the necessary data so that the notebook for the revision plots can be executed quickly.

Also some changes to da_functions. Will need to add the data in the next commit.

In preperation to make the code public for the (hopefully) upcoming paper publication. To many changes to explain easily. The getting started segment has to be revised. And there are lots of semi-redundant functions with _22 and _paper endings. No cleaning up of the inner workings happened. But it is still nice that all the paper plots are not reproducable with one notebook.

Now I am more confused than I was at the beginning

Trying to figure out exactly when things are the same or not

A result of Theresa's visit, this is my first attempt at understanding how the partial analysis increments of Theresa and Martin work using the toymodel. It seems there is still at least one  major bug in my code.

This is a very big change, that now allows having both point and satellite observations. A other very big change is how things are randomized. Now there are specific seed values for all aspects that can contain randomization.Things are still very preliminary, and instead of updating the old functions I created new ones labelled _22. Importantly, the new functions are not backwards compatible! Variance reduction code is also still only preliminary. I also dropped support of the modified shallow water model for now.

The idea is to avoid to stuffing the da_functions file with lots of stuff that doesn't really belong there.

The functions now included are the backbone of version used for the initial paper draft. This was unceremoniasly dumped here before I begin trying to localize the implicit sensitivity approach.

This dramatically helps when localization is applied, but comes with a not well constrained tuning parameter. Will definitely be used as the default method moving forward.

Referred to as quads because the function returns a background, analysis, forecast, and blind/free forecast.

Added the new option to use a LETKF. It is currently much more expensive, but could be made more efficient if needed later on. Getting started now also includes a LETKF example.

Also fixed a bug where the truth was one time step out of sync.

Keep in mind that in comparison to the gaussian the gaspari cohn is slimmer than the gaussian with the same loc_length, so they shouldn't be compared directly against each other.

Now has a little localization section

Changed the various functions needed to include localization in the EnKF. Examples need to be added to the getting started notebook still. And so far only a guassian localization is available.

This is a cleaned up and slightly modified version of the linear advection model I have been playing around with.