"""
Configuration management for the project
"""

import os
from pathlib import Path
from typing import Optional


class Config:
    """
    Central configuration class for the project
    """

    def __init__(self):
        # Project paths
        self.PROJECT_ROOT = Path(__file__).parent.parent.parent
        self.DATA_DIR = self.PROJECT_ROOT / "data"
        self.RAW_DATA_DIR = self.DATA_DIR / "raw"
        self.PROCESSED_DATA_DIR = self.DATA_DIR / "processed"
        self.AUGMENTED_DATA_DIR = self.DATA_DIR / "augmented"
        self.MODELS_DIR = self.DATA_DIR / "models"
        self.LOGS_DIR = self.PROJECT_ROOT / "logs"

        # Create directories if they don't exist
        for directory in [
            self.DATA_DIR,
            self.RAW_DATA_DIR,
            self.PROCESSED_DATA_DIR,
            self.AUGMENTED_DATA_DIR,
            self.MODELS_DIR,
            self.LOGS_DIR,
        ]:
            directory.mkdir(parents=True, exist_ok=True)

        # Image processing settings
        self.TARGET_IMAGE_SIZE = 256  # pixels (256×256)
        self.IMAGE_CHANNELS = 3  # RGB
        self.NORMALIZATION_RANGE = (0, 1)  # Pixel normalization range

        # Data augmentation settings
        self.AUGMENTATION_FACTOR = 5  # Generate 5 variations per image
        self.ROTATION_RANGE = 10  # ±10 degrees
        self.BRIGHTNESS_RANGE = 0.15  # ±15%
        self.ZOOM_RANGE = (0.95, 1.05)  # 95-105%

        # Dataset split settings
        self.TEST_SIZE = 0.2  # 20% for testing
        self.VAL_SIZE = 0.1  # 10% for validation
        self.RANDOM_STATE = 42  # For reproducibility
        self.CV_FOLDS = 5  # Stratified 5-fold cross-validation

        # Deep Learning settings
        self.DL_IMAGE_SIZE = 224  # ResNet50/EfficientNet input
        self.DL_BATCH_SIZE = 8  # Reduced for MPS memory limits with dual backbone
        self.DL_EPOCHS = 100  # More epochs with early stopping
        self.DL_LEARNING_RATE = 1e-4  # Lower LR for fine-tuning backbone
        self.DL_WEIGHT_DECAY = 1e-4
        self.DL_PATIENCE = 15  # Early stopping patience
        self.DL_BACKBONE_FROZEN = True  # Freeze backbone, train only heads (faster)
        self.DL_MODELS_DIR = self.DATA_DIR / "models" / "dl"
        self.DL_EXPANDED_DATA_DIR = self.DATA_DIR / "raw" / "expanded"
        self.DL_EXTERNAL_DATA_DIR = self.DATA_DIR / "raw" / "external"

        # Multi-head model settings (SVDD + classifier heads)
        self.DL_SVDD_EMBEDDING_DIM = 128  # Deep SVDD embedding dimension
        self.DL_HEAD_A_ALPHA = 0.15  # Pokemon classifier (increased: now have non-Pokemon negatives)
        self.DL_HEAD_B_BETA = 0.40   # Back authenticator loss weight
        self.DL_HEAD_C_GAMMA = 0.45  # Front SVDD loss weight (primary mechanism)

        # Training improvements for counterfeit detection
        self.DL_BACK_COUNTERFEIT_WEIGHT = 2.5    # Class weight for counterfeit backs (ratio real/fake: 300/120)
        self.DL_MINORITY_AUGMENT_FACTOR = 2      # Duplication factor for minority class (backs_fake)
        self.DL_CALIBRATION_FBETA = 2.0           # F-beta for threshold calibration (2.0 = recall-weighted)
        self.DL_USE_FOCAL_LOSS = True             # Enable focal loss for Head A/B
        self.DL_FOCAL_GAMMA = 2.0                 # Focal loss gamma (focus on hard examples)
        self.DL_SVDD_CONTRASTIVE_ETA = 1.0        # Weight for contrastive SVDD term (Deep SAD)

        # Ensure DL directories exist
        self.DL_MODELS_DIR.mkdir(parents=True, exist_ok=True)
        self.DL_EXPANDED_DATA_DIR.mkdir(parents=True, exist_ok=True)
        self.DL_EXTERNAL_DATA_DIR.mkdir(parents=True, exist_ok=True)

        # Logging settings
        self.LOG_LEVEL = os.getenv("LOG_LEVEL", "INFO")
        self.LOG_FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
        self.LOG_FILE = self.LOGS_DIR / "cardauth.log"

    def get_dataset_path(self, dataset_type: str = "raw") -> Path:
        """
        Get path to dataset directory

        Args:
            dataset_type: One of 'raw', 'processed', 'augmented'

        Returns:
            Path to the dataset directory
        """
        dataset_map = {
            "raw": self.RAW_DATA_DIR,
            "processed": self.PROCESSED_DATA_DIR,
            "augmented": self.AUGMENTED_DATA_DIR,
        }

        if dataset_type not in dataset_map:
            raise ValueError(
                f"Unknown dataset type: {dataset_type}. "
                f"Choose from: {list(dataset_map.keys())}"
            )

        return dataset_map[dataset_type]

    def get_model_path(self, model_name: str) -> Path:
        """
        Get path to save/load a model

        Args:
            model_name: Name of the model file

        Returns:
            Path to the model file
        """
        return self.MODELS_DIR / model_name

    def __repr__(self) -> str:
        return f"Config(root={self.PROJECT_ROOT})"


# Global config instance
config = Config()