Saving and loading¶

Saving the result¶

Set output_path and run() writes the digital image to disk:

sim = ms.Simulation(..., output_path="result.zarr")
sim.run()

Zarr and NetCDF files keep the dimension names, coordinates (µm), and channel names, and store the full simulation as JSON in the microsim.Simulation attribute. Open them with xarray:

import xarray as xr

result = xr.open_dataarray("result.zarr", engine="zarr")
sim = ms.Simulation.model_validate_json(result.attrs["microsim.Simulation"])

To save intermediate stages, use xarray directly, for example sim.optical_image().to_netcdf("optical.nc").

Saving the simulation¶

A Simulation is a pydantic model, so it can be written to and read from JSON:

from pathlib import Path

Path("sim.json").write_text(sim.model_dump_json(indent=2))

sim = ms.Simulation.model_validate_json(Path("sim.json").read_text())

The JSON is complete: fluorophores loaded from FPbase are saved with their spectra, so loading does not need network access (except for a CosemLabel sample, which queries OpenOrganelle when it is loaded). It also includes the random seed, so loading and running the simulation in the same environment reproduces the result.

You can also write simulations as JSON (or YAML, or anything that loads into a dictionary) by hand. Only the fields you want to change from the defaults are needed:

{
  "truth_space": {"shape": [128, 512, 512], "scale": [0.02, 0.01, 0.01]},
  "output_space": {"downscale": 8},
  "sample": [
    {"distribution": {"type": "matslines", "density": 0.5, "length": 30}, "fluorophore": "EGFP"}
  ],
  "channels": ["wKqWbg::Widefield Green"],
  "modality": {"type": "confocal", "pinhole_au": 0.5},
  "detector": {"camera_type": "CMOS", "qe": 0.8, "read_noise": 1.5},
  "settings": {"random_seed": 42},
  "output_path": "confocal.zarr"
}