Optional ML scoring via ONNX model.

When an ONNX model and companion model_meta.json are provided, each [VariantCall] is scored by a gradient-boosted classifier trained on simulated training data. The resulting probability is stored in [VariantCall::ml_prob].

Compile with --features ml to enable ONNX inference. Without the feature, [MlScorer::load] returns an error and [MlScorer::score] always returns None.

Example

use std::path::Path;
use kam_ml::MlScorer;

let scorer = MlScorer::load(
    Path::new("model.onnx"),
    Path::new("model_meta.json"),
).expect("load model");