GitHub

Ronin.jl

Ronin (Random forest Optimized Nonmeteorological IdentificatioN) removes non-meteorological gates from Doppler radar scans using the machine-learning methodology of Dr. Alex DesRosiers and Dr. Michael Bell. It derives features from raw radar data, trains a Random Forest classifier on them, and applies that model to clean the raw fields in radar scans, with built-in evaluation tooling.

If you use Ronin in published work, please cite DesRosiers and Bell (2023), Artificial Intelligence for the Earth Systems.

Installation

Ronin is in the Julia General Registry:

using Pkg
Pkg.add("Ronin")

Fastest path

The recommended way to train and apply a model is the shipped workflow/ pipeline. From a clone of the repository:

# edit workflow/00_config.jl (experiment name, data paths, model params)
julia --threads=auto workflow/01_split_data.jl   # one-time train/test/val split
julia --threads=auto workflow/02_train.jl        # train the multi-pass cascade
julia --threads=auto workflow/02a_evaluate.jl    # evaluate on the testing set
julia --threads=auto workflow/06_qc.jl           # write QC'd fields

See the Workflow Guide for the full pipeline, the Concepts page for the mental model, and Choosing a QC Entry Point if you are integrating Ronin into your own driver.

What changed in 1.2.0

Version 1.2.0 is the first Julia General Registry release. The documentation below now covers the full current surface; the legacy single-model, hand-tuned-predictor flow is preserved on the Legacy Hand-Tuned Mode page but is no longer the recommended path. Highlights of the 1.1.0 → 1.2.0 delta (see CHANGELOG.md for the authoritative list):

  • Convolution feature mode (task_mode = "convolution") — a kernel bank replaces the hand-crafted predictor file as the recommended feature set.
  • Multi-pass cascade — train_multi_model / composite_prediction / composite_QC, with per-pass met_probs, mask_names, and selected_features.
  • Met-prob-masked convolutions — optional and experimental.
  • Hyperparameter and Pass-2 threshold sweeps — run_hypertuning, sweep_pass2_met_probs.
  • Permutation feature importance — compute_importance.
  • Robust config persistence — save_config / load_config survive future ModelConfig field additions (fixes #34).
  • Workflow scripts — the workflow/ directory and run_workflow.jl orchestrator, with auto-persisted model_config_<EXPERIMENT_NAME>.jld2.
  • Validation-set splitting — split_training_testing_validation!.

Non-breaking deprecations: the v1.1.0 REMOVE_LOW_NCP keyword and ModelConfig.REMOVE_LOW_NCP field still work but emit a deprecation warning; use REMOVE_LOW_SIG_QUALITY. Breaking removals: predict_with_model and the 3-arg evaluate_model(::String, ::String, ::String) are gone — use composite_prediction, run_evaluation, or QC_scan.