src/core/config.py

"""
HITL-KG Configuration Module
Centralized configuration for the entire system.
"""

import os
from dataclasses import dataclass, field
from typing import Dict, List, Optional, Any
from pathlib import Path
import yaml
import logging

logger = logging.getLogger(__name__)


@dataclass
class EmbeddingConfig:
    """Configuration for embedding-based search."""
    model_name: str = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
    cache_dir: str = "./data/embeddings"
    dimension: int = 384
    batch_size: int = 32
    device: str = "cpu"  # "cpu", "cuda", "mps"


@dataclass
class DatasetConfig:
    """Configuration for a single dataset."""
    name: str
    source_type: str  # "obo", "owl", "csv", "json"
    source_url: Optional[str] = None
    source_path: Optional[str] = None
    entity_category: str = "entity"
    cache_enabled: bool = True
    cache_max_age_days: int = 7


@dataclass
class LLMConfig:
    """Configuration for LLM providers."""
    provider: str = "local"  # "openai", "local"
    model: str = "gpt-4o-mini"
    temperature: float = 0.7
    max_tokens: int = 2048
    api_key: Optional[str] = None


@dataclass
class AppConfig:
    """Main application configuration."""
    # Paths
    data_dir: Path = field(default_factory=lambda: Path("./data"))
    cache_dir: Path = field(default_factory=lambda: Path("./data/cache"))
    sessions_dir: Path = field(default_factory=lambda: Path("./data/sessions"))
    
    # Server
    host: str = "0.0.0.0"
    port: int = 7860
    debug: bool = False
    
    # Sessions
    session_max_age_hours: int = 24
    session_cleanup_interval_minutes: int = 5
    max_sessions: int = 1000
    
    # Embeddings
    embedding: EmbeddingConfig = field(default_factory=EmbeddingConfig)
    
    # LLM
    llm: LLMConfig = field(default_factory=LLMConfig)
    
    # Datasets
    datasets: List[DatasetConfig] = field(default_factory=list)
    
    # UI
    default_language: str = "en"
    supported_languages: List[str] = field(
        default_factory=lambda: ["en", "uk", "ru", "es", "de", "fr"]
    )
    
    def __post_init__(self):
        """Ensure directories exist."""
        for path in [self.data_dir, self.cache_dir, self.sessions_dir]:
            path.mkdir(parents=True, exist_ok=True)


# Default medical datasets configuration
DEFAULT_MEDICAL_DATASETS = [
    DatasetConfig(
        name="disease_ontology",
        source_type="obo",
        source_url="https://raw.githubusercontent.com/DiseaseOntology/HumanDiseaseOntology/main/src/ontology/doid.obo",
        entity_category="disease",
    ),
    DatasetConfig(
        name="symptom_ontology", 
        source_type="obo",
        source_url="https://raw.githubusercontent.com/DiseaseOntology/SymptomOntology/main/symp.obo",
        entity_category="symptom",
    ),
]


def load_config(config_path: Optional[str] = None) -> AppConfig:
    """
    Load configuration from file or use defaults.
    
    Priority:
    1. Explicit config file path
    2. Environment variable HITL_KG_CONFIG
    3. ./config.yaml if exists
    4. Default configuration
    """
    config_file = None
    
    if config_path:
        config_file = Path(config_path)
    elif os.environ.get("HITL_KG_CONFIG"):
        config_file = Path(os.environ["HITL_KG_CONFIG"])
    elif Path("./config.yaml").exists():
        config_file = Path("./config.yaml")
    
    if config_file and config_file.exists():
        try:
            with open(config_file) as f:
                data = yaml.safe_load(f)
            return _parse_config(data)
        except Exception as e:
            logger.warning(f"Failed to load config from {config_file}: {e}")
    
    # Return default config with medical datasets
    config = AppConfig()
    config.datasets = DEFAULT_MEDICAL_DATASETS
    
    # Override from environment
    if os.environ.get("OPENAI_API_KEY"):
        config.llm.api_key = os.environ["OPENAI_API_KEY"]
        config.llm.provider = "openai"
    
    if os.environ.get("PORT"):
        config.port = int(os.environ["PORT"])
    
    return config


def _parse_config(data: Dict[str, Any]) -> AppConfig:
    """Parse configuration dictionary into AppConfig."""
    config = AppConfig()
    
    # Parse simple fields
    for key in ["host", "port", "debug", "default_language"]:
        if key in data:
            setattr(config, key, data[key])
    
    # Parse paths
    for key in ["data_dir", "cache_dir", "sessions_dir"]:
        if key in data:
            setattr(config, key, Path(data[key]))
    
    # Parse embedding config
    if "embedding" in data:
        config.embedding = EmbeddingConfig(**data["embedding"])
    
    # Parse LLM config
    if "llm" in data:
        config.llm = LLMConfig(**data["llm"])
    
    # Parse datasets
    if "datasets" in data:
        config.datasets = [DatasetConfig(**ds) for ds in data["datasets"]]
    else:
        config.datasets = DEFAULT_MEDICAL_DATASETS
    
    return config


# Global config instance (lazy loaded)
_config: Optional[AppConfig] = None


def get_config() -> AppConfig:
    """Get the global configuration instance."""
    global _config
    if _config is None:
        _config = load_config()
    return _config