config/settings.yaml

data:
  train_path: data/twitter.csv
  text_column: text
  target_column: sentiment
  class_order: [negative, neutral, positive]

model:
  artifact_dir: artifacts
  pipeline_filename: sentiment_pipeline.joblib
  test_size: 0.2
  random_state: 42
  max_features: 5000
  ngram_range: [1, 2]
  class_weight: balanced

training:
  cv_folds: 5
  scoring: f1_macro