Skip to contents

Normalise and validate a config for the R transmission engine

Source: R/sim_params.R
sim_params.Rd

Python's params.py is 1,009 lines, and it is tempting to call it "just coercion" and drop it – but that is only safe if the engine's sole input is make_simulation_config() output, and it is not: the engine also accepts raw lists and file paths. Scalar-to-matrix broadcasting, dimension checking and non-finite rejection all live there today and have to live somewhere afterwards. This is that somewhere.

Usage

sim_params(config, components = SIM_PIPELINE, mode = c("rng", "replay"))

Arguments

config

Config list, or a path to a .json / .json.gz file.

components

Pipeline subset that will be run; determines which compartments Census sums and which parameters are required.

mode

"rng" (production) or "replay" (the laser-cholera parity harness); decides which mortality inputs are read.

Value

A list of validated engine parameters, including the config as supplied (a file path is parsed, a list is kept unchanged) under $config.

Details

Orientation is the single most dangerous thing here. Configs carry time-varying fields as [npatches][nticks] (location-major, matching the JSON on disk), while the engine's internal state is [nticks + 1, npatches] (time-major, matching Python's frames). This function transposes once, at the boundary, and every dimension is asserted rather than assumed – a silently transposed input is the likeliest way to get plausible-looking wrong answers out of the whole port.