baseline.json

{
  "task": "gsw-vs-de language identification (binary LID)",
  "baseline": "char n-gram multinomial Naive Bayes",
  "config": {
    "ngram_range": [
      2,
      4
    ],
    "alpha": 1.0,
    "vocab_size": 81001,
    "deterministic": "closed-form counts; no RNG/seed"
  },
  "train_n": 468,
  "scorecard": {
    "n": 200,
    "accuracy": 0.67,
    "macro_f1": 0.6323529411764706,
    "per_class_f1": {
      "de": 0.5147058823529411,
      "gsw": 0.75
    }
  },
  "command": "PYTHONPATH=src .venv/bin/python scripts/run_baseline.py"
}