# Machine Learning Model Configuration

pattern_graders:
  # Individual pattern grading models
  base_model_type: "RandomForestClassifier"  # RandomForestClassifier, XGBoostClassifier
  
  random_forest:
    n_estimators: 100
    max_depth: 10
    min_samples_split: 5
    min_samples_leaf: 2
    max_features: "sqrt"
    random_state: 42
    n_jobs: -1
  
  xgboost:
    n_estimators: 100
    max_depth: 6
    learning_rate: 0.1
    subsample: 0.8
    colsample_bytree: 0.8
    random_state: 42
  
  # Feature selection
  feature_selection:
    enabled: true
    method: "mutual_info"  # mutual_info, f_test, chi2
    top_k_features: 50
  
  # Training configuration
  training:
    cv_folds: 5
    scoring_metric: "f1_weighted"
    early_stopping: true
    patience: 10

setup_classifier:
  # Meta-model for complete setup classification
  model_type: "RandomForestClassifier"
  
  # Strategy types
  strategies:
    continuation:
      time_window_start: "03:00"
      time_window_end: "03:15"
      min_pattern_count: 2
      required_patterns: ["fvg", "order_block"]
    
    reversal:
      time_window_start: "03:30"
      time_window_end: "03:50"
      min_pattern_count: 2
      required_patterns: ["fvg", "liquidity"]

# Model evaluation
evaluation:
  metrics:
    - "accuracy"
    - "precision"
    - "recall"
    - "f1_score"
    - "roc_auc"
  
  min_accuracy_threshold: 0.75
  min_precision_threshold: 0.70
  min_recall_threshold: 0.65

# Hyperparameter tuning
tuning:
  enabled: false
  method: "grid_search"  # grid_search, random_search, optuna
  n_iter: 50
  cv_folds: 5
  
  grid_search:
    param_grids:
      random_forest:
        n_estimators: [50, 100, 200]
        max_depth: [5, 10, 15]
        min_samples_split: [2, 5, 10]

# Model registry
registry:
  track_versions: true
  auto_promote: false  # Auto-promote best model to "latest"
  min_improvement: 0.02  # Minimum improvement to promote (2%)