# --- Data Parameters ---
data:
  db_dir: '../data/crypto_market_data' # Path to the directory containing the market data database (relative to where main.py is run).
  exchange: 'bnbspot'                    # Name of the exchange table/data source in the database.
  ticker: 'SOL-USDT'                       # Instrument identifier (e.g., trading pair) within the exchange data.
  vol_window: 30                      # Task 1.1: Window size for volatility calculation
  vol_quantile: 0.5                   # Task 1.1: Keep samples where vol > this quantile
  label_smoothing: 0.05               # Task 1.2 / Revision 2-B: Apply label smoothing to binary targets (0.0 = off, 0.1 = [0.05, 0.95])

# --- Feature Engineering Params --- 
features:
  # Placeholder for potential future config, like VIF skip - Task 2.5
  # skip_vif: false
  # Define minimal whitelist here? Or keep in feature_engineer.py fallback?
  minimal_whitelist: [
      "return_1m", "return_15m", "return_60m", "ATR_14", "volatility_14d",
      "chaikin_AD_10", "svi_10", "EMA_10", "EMA_50",
      "hour_sin", "hour_cos", "week_sin", "week_cos", # Added week cycle
      # Add new micro-structure features if they prove useful and pass selection
      "spread_proxy", "vol_norm_volume_spike", "return_asymmetry", 
      "close_location_value", "keltner_band_pos"
  ]

# --- Data Split (Walk-Forward Settings - Replaces split_ratios) --- #
walk_forward:
  enabled: true             # Enable walk-forward validation loop
  train_days: 60           # Duration of the training set in days for each fold
  val_days: 14             # Duration of the validation set in days for each fold
  test_days: 14            # Duration of the test set in days for each fold
  step_days: 14            # How many days to step forward for the next fold (usually val_days or test_days)
  initial_offset_days: 0 # Optional: Skip first N days of data before starting first fold

# --- GRU Model Parameters --- #
gru:
  # General
  prediction_horizon: 5               # How many steps ahead the model predicts.
  lookback: 60                        # Sequence length input to the GRU.
  # --- New Label/Version Params (v3 Rev) --- #
  use_ternary: true                   # Task 1.3 / Revision 7: Use ternary (up/flat/down) labels instead of binary.
  flat_sigma_multiplier: 0.3          # Task 1.3 / Revision 1-C / Revision 7: k for ternary flat threshold (eps = k * rolling_sigma_N).
  # --- v2 Specific Params (Legacy) --- #

# --- GRU v3 Model Specific Parameters --- #
gru_v3:
  # ... existing gru_v3 params ...
  gru_units: 96                      # Hidden units in the GRU layer
  attention_units: 16                # Attention mechanism units (0 disables attention)
  learning_rate: 1e-4                # Learning rate for Adam optimizer
  focal_gamma: 2.0                   # Gamma parameter for focal loss
  focal_label_smoothing: 0.1         # Label smoothing for focal loss
  huber_delta: 1.0                   # Delta parameter for huber loss
  loss_weight_mu: 0.3                # Weight for mu output in loss
  loss_weight_dir3: 1.0              # Weight for dir3 output in loss
  l2_reg: 1e-4                       # L2 regularization weight

# --- Hyperparameter Tuning --- #
hyperparameter_tuning:
  gru:
    sweep_enabled: false             # Set to true to enable hyperparameter tuning with Optuna
    sweep_n_trials: 20               # Number of trials for the Optuna study
    sweep_timeout: 7200              # Maximum time in seconds for the hyperparameter search
    enable_pruning: true             # Enable trial pruning for efficiency
    pruning_warmup_trials: 5         # Number of trials before pruning is applied

# --- Calibration Parameters --- #
calibration:
  method: 'vector'                  # Calibration method ('temperature' or 'vector')
  edge_threshold: 0.2             # Default probability edge threshold (can be optimized per fold)
  optimize_edge_threshold: true     # If true, optimize edge_threshold per fold using Youden's J
  # Rolling Calibration (Implemented in Backtester)
  rolling_enabled: false           # Enable rolling calibration during backtest
  recalibrate_every_n: 5000        # How often (in steps/bars) to refit the calibrator during backtest
  recalibration_window: 20000      # How many recent steps/bars to use for refitting
  coverage_alarm_enabled: false    # Enable the coverage alarm to trigger early recalibration
  coverage_alarm_threshold_drop: 0.03 # Trigger alarm if hit rate < edge_thr - drop
  coverage_alarm_window: 1000      # Monitor hit rate over this many bars for the alarm

# --- Validation Gates --- #
validation_gates:
  baseline:
    ci_lower_bound_threshold: 0.52 # Min 95% CI lower bound for logistic baseline accuracy
  gru:
    edge_filtered_acc_ci_lower_threshold: 0.62 # Revision 6: Updated threshold
    brier_score_threshold: 0.22              # Revision 6: Updated threshold (Max allowed Brier score)
    edge_filtered_min_samples: 30            # Min samples required for edge-filtered CI calculation
  backtest:
    # sharpe_threshold: 1.0 # Revision 6: Removed - Replaced by median check in final_release_decision
    max_drawdown_threshold: 15.0           # Max allowed drawdown percentage
    win_rate_threshold: 55.0               # Min win rate percentage
    # Note: Median Sharpe threshold (e.g., >= 1.3 for >= 5 folds) is applied in final_release_decision

# --- SAC Agent Parameters --- #
sac:
  # ... existing sac params ...
  use_per: false                     # Enable Prioritized Experience Replay (Requires PER implemented in SACTrainer)
  # ... rest of sac params ...

# --- SAC Aggregation (Post-Processing) --- #
sac_aggregation:
  enabled: false                   # Enable aggregation/merging of SAC agents from folds
  method: 'average_weights'        # Strategy: 'average_weights', 'voting_ensemble' (Requires separate implementation)

# --- Environment Parameters (Used by train_sac.py & backtester.py) --- #
environment:
  initial_capital: 10000.0            # Notional capital for env/backtest consistency.
  transaction_cost: 0.0005            # Fractional cost per trade (e.g., 0.0005 = 0.05%).
  # --- New Env Params (v3 Rev) --- #
  reward_scale: 100.0                 # Task 5.1 / Revision 4-A / Revision 7: Multiplier applied to the raw environment reward.
  action_penalty_lambda: 0.005        # Task 5.4 / Revision 4-A / Revision 7: Coefficient (lambda) for action magnitude penalty (reward -= lambda * action^2).

# ... existing code ...