Might as well reuse https://github.com/ali-ramadhan/antarctic-siz-stress/blob/master/logging.ini

Right now everything is written to uncompressed pickle files. At the very least, the pickle files should be compressed. Would be good to use NetCDF as it's a familiar format.

Unfortunately last time I tried, Parcels' NetCDF output didn't work when running parallel jobs with joblib so I had to use joblib pickles. If this is still the case, maybe all the pickle files can be merged into a massive NetCDF file at the end?

The interactions might create Lagrangian microbes with properties such as species (integer or string) or genome (binary string). These should still fit in NetCDF files.

Should be able to:

• Plot the position and species of the microbes at each time step.
• Make movies? Should I use ffmpeg-python?

Some examples:

• Species
• Genome or genetic data (binary string)

Pinging: @gregbritten

Right now the InteractionSimulator contains a ParticleAdvecter which isn't good because we can't run multiple interaction simulations by themselves, they need to be tied to a ParticleAdvecter at run time.

They should be decoupled!

See:

• https://packaging.python.org/tutorials/packaging-projects/
• https://docs.python-guide.org/writing/structure/
• https://github.com/kennethreitz/samplemod

Right now pairs are queried from the kd-tree using a radius that is in "degrees^2" (units of lat*lon) essentially but it might be nice to specify it in meters which is then converted to "degrees^2".

Once we have modules this should be easy to implement so it's super easy to restart.

Possible choices:

• Uniformly arranged in a lat/lon box.
• Gaussian distribution with some mean (lat, lon) and standard deviation (sigma_lat, sigma_lon).

Would be nice to have a zero velocity field for debugging purposes. Also for turning off advection.

To show off the capabilities of the package!

Since I'll be refactoring this package, might as well switch to the brand new Parcels 2.0 (still in beta).

See Parcels releases for details on breaking changes: https://github.com/OceanParcels/parcels/releases

Should have some examples for potential users:

• Simple barebones single-core simulation (show the entire pipeline)
• Large multi-core simulation with OSCAR fields

Run for a full year:

• Vary # of particles: 10,000 100,000 1,000,000
• Vary diffusivity: 0, low κ (fuzzy motion, Kh=20), medium κ (Kh=100), large κ (get kicked out of eddies, Kh=500)
• Vary the interaction: netural (change p=0.5, 0.7, 0.9, 0.999, 1), non-neutral (pRS=0.501, 0.51, 0.6, 0.7, 0.99)

Cannot plot multiple frames at once...

I suspect it's because of xarray accessing the data over OpeNDAP...

Should be able to:

• Keep track of arbitrary microbe properties.
• Accept a user-defined interaction between two Lagrangian microbes of the form `interaction(microbe1, microbe2).
• Maybe also a self-interaction of the form self_interaction(microbe).
• Also NetCDF output (preferably).

Should be able to:

• Choose a velocity field (OSCAR, LLC4320, chaotic mixing, etc.)
• Choose number of processors to run on.
• Choose number of particles and their initial distribution.
• Choose time step, simulation length, output frequency, etc.
• Writes output to NetCDF preferably.
• Changes the velocity field as time progresses (especially relevant for OSCAR and LLV4320).

Right now it's impractical to run on 20+ cores as each core will download the OSCAR dataset which causes network issues and gets our connections rejected from NASA's servers.

A simple solution would be to just download and save the OSCAR dataset to disk. Then 100+ cores can read it.

This should also resolve #23.

Gotta pick a favorite python testing framework and write unit tests along the way.

Right now joblib does just fine for multi-core parallelization but for distributed computing might be good to check out Dask: https://docs.dask.org/en/latest/

We'll probably need this if we want to do >10,000,000 particles.

Would be nice to refactor the analysis scripts into nice modules or functions. Right now they are:

• Species count time series.
• Pairwise haversine distance histogram (in Julia).