What is the best approach to load external realistic layers into covasim?

[rlatkows]

Dear all,

I really appreciate publishing tutorials along with version 2.0. It helps a lot to understand some parameters much better than just reading tests. Still I have some questions regarding Layers.

I’m extending covasim with some additional properties of the agents like administrative district hierarchy or connection networks based on [almost] real data.

What is suggested architectonic way to populate realistic size layers (e.g. 1e6 students with 20e6 graph edges)?

• use add_layers functionality with specifying all pairs explicitly (clean)
• create new method like make_hybrid_contacts (chosen due to lack of clarity in parameters before publishing tutorials)
• directly populate layer dictionaries from database elsewhere

I don’t afraid so much of RAM (the RAM is strong with this one environment I have), but rather execution speed and maintaining compatibility with next versions of covasim.

Was it also considered to change the data model of Layer class from direct pair representation to a dict like p1: [list of all p2], and move the data integrity part (symmetricity [and/or transitive closure for some layers]) for data preparation phase? Or rather in contrary, did someone observed that numba optimization of paired connections work better?

Best regards,
Rafal