Roadmap

xarray-simlab's roadmap

v0.1.0

• Provide a small framework for creating computational models (Model objects) from re-usable model components (Process subclasses). Each Process subclass:

  • has a declarative interface that defines a set of variables (Variable, ForeignVariable, VariableList, VariableGroup objects) as class attributes ;
  • implements specific methods like .initialize(), .run_step(), .finalize_step() and/or finalize(), which use/set values for the variables declared in the Process.

• Provide an xarray extension (Dataset.simlab accessor) that allows:

  • Creating new data variables and/or coordinates for the inputs and time steps of a given Model ;
  • Run the model using the ds.simlab.run() method that returns a new Dataset with both model inputs and outputs.

• Write tests and documentation.

Next

• For large simulations, save model outputs (snapshots) to disk as the simulation proceed and then return the new Dataset as an open file (currently all snapshots are kept in memory).
• Logging and progress status of a simulation. Maybe use the logging module (add a logger that can be used in each process).
• Allow the processes in a Model to be run in parallel (using Dask?) at each time step during a simulation, and also during initialization and finalization. This should be quite straightforward as the processes in a Model and their inter-dependencies together form a Directed Acyclic Graph (DAG).
• Run multiple models in parallel (using dask?) with different input values, using a single Dataset object. This may be simply achieved using a new method that might look like ds.simlab.run_multi('input_values_dim') where input_values_dim is a multi-index dimension that represents all combinations of model input values.
• Provide a command-line interface to run models and also to automatically generate input files (use YAML, toml or other formats).