Quickstart

Install

pip install maldet[lightning,mlflow]

Scaffold a detector

maldet scaffold --template rf --name mydet --out ./mydet
cd ./mydet
pip install -e .
maldet check
maldet describe

Write your features

Open src/mydet/features.py. Replace the ELF parse with your feature logic. The class must satisfy the FeatureExtractor protocol:

class MyFeatures:
    output_shape = (128,)
    dtype = "float32"
    def extract(self, sample): ...

Train locally

Write a config.yaml:

defaults: [_self_]
stage: train
paths:
  config_dir: .
  output_dir: ./output
  samples_root: ./samples
  source_model: ./output/model
data:
  train_csv: ./train.csv
  test_csv: ./test.csv
  predict_csv: ./predict.csv
model:
  _target_: mydet.models.make_rf
  n_estimators: 100

Run:

maldet run train --config config.yaml

Artifacts land under ./output/ (model/, events.jsonl, metrics.json, manifest.json).