Adding a model

A model is a type that implements phytkit_core::Model. models/bta.rs is a short example to copy.

1. Implement the trait

use crate::model::{col, get, param, Columns, Model, ParamSpec, Params};

pub struct MyModel { params: Vec<ParamSpec> }

impl MyModel {
    pub fn new() -> Self {
        MyModel {
            params: vec![
                param("a", 1.0, 0.0, 10.0).describe("scale"),
                param("b", 0.5, 0.0, 5.0),
                param("c", 0.0, -1.0, 1.0).fixed(),   // held fixed unless freed with --fit
            ],
        }
    }
}

impl Model for MyModel {
    fn name(&self) -> &str { "myModel" }
    fn inputs(&self) -> &[&'static str] { &["Q"] }
    fn params(&self) -> &[ParamSpec] { &self.params }
    fn predict(&self, p: &Params, x: &Columns) -> Vec<f64> {
        // Read parameters once, outside the loop: looking them up per point is slow.
        let (a, b, c) = (get(p, "a"), get(p, "b"), get(p, "c"));
        col(x, "Q").iter().map(|q| a * q / (q + b) + c).collect()
    }
}

Optional methods:

Method	Use it for
jacobian and has_jacobian	Exact derivatives, instead of finite differences
penalties and has_penalties, constraints	Soft constraints
derived	Values computed from the parameters, reported with each fit (like psi_tlp)
latex	The equation, shown by pk describe and in these docs

2. Register it

Add the module to models/mod.rs, and add an instance to builtin_models(). The model is then available to pk describe, fit_arrays, fit_many and predict.

3. Add it to a tool (optional)

To fit it from files with pk fit, add a ToolModel to a tool in crates/cli/src/tools.rs. That entry sets:

• the label,
• the parser that reads its inputs from a file, and
• an optional seed function that sets starting values from the data.

4. Test and document it

Add tests in the model's file, run python3 docs/gen_models.py to add it to Models and parameters, and describe it in Concepts.