🧩🦀 why-rs: A Causal Inference library for Rust

Usage

To use the crate, you can include it in your cargo.toml:

[dependencies]
why_rs = "0.2.1"

Or, get the latest version directly from github:

[dependencies]
my_crate = { git = "https://github.com/jwhogg/why-rs.git" }

Example usage

    let dag: DAG = dag!( //macro for defining DAGs similar to DOT files
        "A" => "C",
        "B" => "C",
        "C" => "D"
    );

    let mut fcm = FCM::from_dag(dag);

    let df = generate_data(500); //simulate some practice data

    //fit the LR mechanism given training data
    let mut c_mechanism = LinearRegression::new();
    c_mechanism.fit(df.clone(), Variable::from("C"), &fcm);

    //there is also the option to manually define coefficients:
    let noise_d = rand::thread_rng().gen_range(0..5);
    let mut d_mechanism = LinearRegression::from(vec![0.8], 1, noise_d)

    fcm = fcm //for this to actually run, you would need to define rules for A and B- see examples/linear_regression
        .rule("C", c_mechanism)
        .rule("D", d_mechanism);

    let result = fcm.sample(15); //sample 15 rows
    println!("normal sampling: {}", result);

    let intervened_df = fcm.interventional_samples(
    intervene!('B': 0), // The Macro in action
    5 // n_samples
    );

    println!("Intervened Graph: {}", intervened_df);

Running examples

Currently, there is support for the DAG and FCM (Functional Causal Model) data types. The following functionality is supported:

• Parsing a .dot file
• Sampling from an FCM
• Causal Discovery with the PC algorithm
• Intervening on a FCM
• Plugging-in custom models for FCM mechanisms

Try it yourself:

First, clone the repository: git clone https://github.com/jwhogg/why-rs.git cd why-rs

Sampling:

cargo run --example fcm_test

PC algorithm:

cargo run --example pc_test

Interventions:

cargo run --example intervention

Parsing .DOT files:

cargo run --example io_test