Model definition format

For an example model definition file, see examples/h3n2-model.yaml.

TreeFlow’s command line interfaces use a YAML model definition format. Each model definition file has four sections:

tree:
    ...
clock:
    ...
site:
    ...
substitution:
    ...

Each section takes a YAML mapping from the name for the selection of that model component to its parameters.

Each parameter can be a fixed value, or a prior distribution if the parameter is to be estimated.

Prior distributions are specified as mappings from a distribution name to parameters.

For example:

substitution:
    hky:
        kappa:
            2.0
        frequencies:
            dirichlet:
                concentration: [2.0, 2.0, 2.0, 2.0]

Documentation on all the options is still to come, see the source code at treeflow.model.phylo_model.phylo_model_to_joint_distribution for reference for now.

Prior distributions

For parameters, see the corresponding TensorFlow Probability distribution.