The TOML parameter file interface

The complete user interface consists of two files in TOML format

1. A user-defined experiment file - in the local experiment directory
2. A defaults file - in src/ directory of CLIMAParameters.jl

Parameter style-guide

A parameter is determined by its unique name. It has possible attributes

1. value
2. type
3. description
4. prior
5. tag
6. transformation

Minimal parameter requirement to run in CliMA

[molar_mass_dry_air]
value = 0.03
type = "float"

A more informative parameter (e.g. found in the defaults file)

[molar_mass_dry_air]
value = 0.02897
type = "float"
description = "Molecular weight dry air (kg/mol)"

Properly tagged parameter

To add a tag to a parameter, set the tag field with a list of tags. Tags are an optional convenience and do not create a namespace. All parameter names must be unique.

As an initial convention, parameters will be tagged with the component(s) in which they are used. This convention will be changed as we see how packages use tags.

[prandtl_number_0_grachev]
value = 0.98
type = "float"
description = "Pr_0 for Grachev universal functions. From Grachev et al, 2007. DOI: 10.1007/s10546-007-9177-6"
tag = ["SurfaceFluxes"]

If this convention is followed, to obtain the parameters used to build tagged by "surfacefluxes", one could call for example:

surfacefluxes_params = get_tagged_parameter_values(toml_dict, "surfacefluxes")

To match tags, punctuation and capitalization is removed. For more information, see fuzzy_match.

A more complex parameter for calibration

[neural_net_entrainment]
value = [0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0]
type = "float"
description = "NN weights to represent the non-dimensional entrainment function"
prior = "MvNormal(0,I)"

Interaction of the files

On read an experiment file, the default file is also read and any duplicate parameter attributes are overwritten e.g. If the minimal example above was loaded from an experiment file, and the informative example above was in the defaults file, then the loaded parameter would look as follows:

[molar_mass_dry_air]
value = 0.03
type = "float"
description = "Molecular weight dry air (kg/mol)"

Here, the value field has been overwritten by the experiment value.

File and parameter interaction with CliMA

CLIMAParameters.jl provides several methods to parse, merge, and log parameter information.

Loading from file

We provide the following methods to load parameters from file

create_toml_dict(Float64; override_filepath, default_filepath)
create_toml_dict(Float64; override_filepath)

The Float64 (or Float32) defines the requested precision of the returned parameters.

Typical usage involves passing the local parameter file

import CLIMAParameters
local_experiment_file = joinpath(@__DIR__,"local_exp_parameters.toml")
toml_dict = CLIMAParameters.create_toml_dict(; override_file = local_experiment_file)

If no file is passed it will use only the defaults from CLIMAParameters.jl (causing errors if required parameters are not within this list).

You can also pass Julia Dicts directly to override_filepath and default_filepath.

If you want to use more than two TOML files, you can merge them with merge_toml_files(filepaths...). By default, duplicate TOML entries are not allowed, but this can be changed by setting override = true.

The parameter dict is then used to build the codebase (see Parameter Retrieval for usage and examples).

Logging parameters

Once the CliMA components are built, it is important to log the parameters. We provide the following method

log_parameter_information(toml_dict, filepath; strict=false)

Typical usage will be after building components and before running

import Thermodynamics
therm_params = Thermodynamics.ThermodynamicsParameters(toml_dict)
#... build(thermodynamics model,therm_params)

log_file = joinpath(@__DIR__,"parameter_log.toml")
CLIMAParameters.log_parameter_information(toml_dict,log_file)

# ... run(thermodynamics_model)

This function performs two tasks

1. It writes a parameter log file to log_file.
2. It performs parameter sanity checks.

Continuing our previous example, imagine molar_mass_dry_air was extracted in ThermodynamicsParameters. Then the log file will contain:

[molar_mass_dry_air]
value = 0.03
type = "float"
description = "Molecular weight dry air (kg/mol)"
used_in = ["Thermodynamics"]

The additional attribute used_in displays every CliMA component that used this parameter.