HITL-KG / src /core /dataset_loader.py
avojarot's picture
Upload 22 files
c5880fb verified
"""
Dataset Loader Module (Refactored)
Generic dataset loading supporting multiple formats:
- OBO (Open Biomedical Ontologies)
- CSV/TSV
- JSON/JSON-LD
- Custom adapters
Configuration-driven to support any domain, not just medical.
"""
import os
import re
import json
import csv
import logging
import hashlib
import urllib.request
from pathlib import Path
from abc import ABC, abstractmethod
from datetime import datetime
from typing import Dict, List, Optional, Tuple, Any, Type
from dataclasses import dataclass, field
from .knowledge_graph import Entity, EntityCategory, KnowledgeGraph
from .config import DatasetConfig, get_config
logger = logging.getLogger(__name__)
@dataclass
class OntologyTerm:
"""Generic ontology term representation."""
id: str
name: str
definition: str = ""
synonyms: List[str] = field(default_factory=list)
xrefs: Dict[str, str] = field(default_factory=dict)
is_a: List[str] = field(default_factory=list)
relationships: List[Tuple[str, str]] = field(default_factory=list)
namespace: str = ""
is_obsolete: bool = False
def to_entity(self, category: EntityCategory) -> Entity:
"""Convert to Entity."""
return Entity(
id=self.id,
name=self.name,
category=category,
description=self.definition,
synonyms=self.synonyms,
xrefs=self.xrefs,
properties={"is_a": self.is_a, "namespace": self.namespace}
)
class DatasetAdapter(ABC):
"""Abstract base class for dataset adapters."""
@abstractmethod
def parse(self, content: str) -> Dict[str, OntologyTerm]:
"""Parse content and return dictionary of terms."""
pass
@abstractmethod
def can_handle(self, source_type: str) -> bool:
"""Check if this adapter can handle the source type."""
pass
class OBOAdapter(DatasetAdapter):
"""Parser for OBO (Open Biomedical Ontologies) format."""
def can_handle(self, source_type: str) -> bool:
return source_type.lower() == "obo"
def parse(self, content: str) -> Dict[str, OntologyTerm]:
"""Parse OBO format content."""
terms = {}
# Split into stanzas
stanzas = re.split(r'\n\[', content)
for stanza in stanzas[1:]: # Skip header
if stanza.startswith('Term]'):
term = self._parse_term(stanza[5:])
if term and not term.is_obsolete:
terms[term.id] = term
logger.info(f"Parsed {len(terms)} terms from OBO content")
return terms
def _parse_term(self, stanza: str) -> Optional[OntologyTerm]:
"""Parse a single term stanza."""
data = {
"id": "", "name": "", "definition": "",
"synonyms": [], "xrefs": {}, "is_a": [],
"relationships": [], "namespace": "", "is_obsolete": False
}
for line in stanza.split('\n'):
line = line.strip()
if not line or line.startswith('!') or ':' not in line:
continue
tag, _, value = line.partition(':')
tag, value = tag.strip(), value.strip()
if tag == 'id':
data['id'] = value
elif tag == 'name':
data['name'] = value
elif tag == 'def':
match = re.match(r'"([^"]*)"', value)
if match:
data['definition'] = match.group(1)
elif tag == 'synonym':
match = re.match(r'"([^"]*)"', value)
if match:
data['synonyms'].append(match.group(1))
elif tag == 'xref':
if ':' in value:
xref_ns, _, xref_id = value.partition(':')
xref_id = xref_id.split()[0] if ' ' in xref_id else xref_id
data['xrefs'][xref_ns.strip()] = xref_id.strip()
elif tag == 'is_a':
parent_id = value.split('!')[0].strip()
data['is_a'].append(parent_id)
elif tag == 'relationship':
parts = value.split()
if len(parts) >= 2:
data['relationships'].append((parts[0], parts[1]))
elif tag == 'is_obsolete':
data['is_obsolete'] = value.lower() == 'true'
elif tag == 'namespace':
data['namespace'] = value
if data['id'] and data['name']:
return OntologyTerm(**data)
return None
class CSVAdapter(DatasetAdapter):
"""Parser for CSV/TSV format datasets."""
# Default column mappings
DEFAULT_MAPPINGS = {
"id": ["id", "ID", "identifier", "code"],
"name": ["name", "Name", "label", "Label", "title"],
"definition": ["definition", "description", "Description", "desc"],
"synonyms": ["synonyms", "aliases", "alt_names"],
}
def __init__(self, column_mappings: Optional[Dict[str, str]] = None):
self.column_mappings = column_mappings or {}
def can_handle(self, source_type: str) -> bool:
return source_type.lower() in ["csv", "tsv"]
def parse(self, content: str) -> Dict[str, OntologyTerm]:
"""Parse CSV content."""
terms = {}
# Detect delimiter
dialect = csv.Sniffer().sniff(content[:1024])
reader = csv.DictReader(content.splitlines(), dialect=dialect)
# Map columns
col_map = self._map_columns(reader.fieldnames or [])
for row in reader:
term = self._row_to_term(row, col_map)
if term:
terms[term.id] = term
logger.info(f"Parsed {len(terms)} terms from CSV content")
return terms
def _map_columns(self, fieldnames: List[str]) -> Dict[str, str]:
"""Map fieldnames to standard term fields."""
col_map = {}
for field, possible_names in self.DEFAULT_MAPPINGS.items():
# Check explicit mappings first
if field in self.column_mappings:
col_map[field] = self.column_mappings[field]
else:
# Try to auto-detect
for name in possible_names:
if name in fieldnames:
col_map[field] = name
break
return col_map
def _row_to_term(self, row: Dict, col_map: Dict[str, str]) -> Optional[OntologyTerm]:
"""Convert CSV row to OntologyTerm."""
term_id = row.get(col_map.get("id", ""), "")
name = row.get(col_map.get("name", ""), "")
if not term_id or not name:
return None
definition = row.get(col_map.get("definition", ""), "")
# Parse synonyms (comma-separated or JSON array)
synonyms_raw = row.get(col_map.get("synonyms", ""), "")
if synonyms_raw.startswith("["):
try:
synonyms = json.loads(synonyms_raw)
except json.JSONDecodeError:
synonyms = []
else:
synonyms = [s.strip() for s in synonyms_raw.split(",") if s.strip()]
return OntologyTerm(
id=term_id,
name=name,
definition=definition,
synonyms=synonyms
)
class JSONAdapter(DatasetAdapter):
"""Parser for JSON format datasets."""
def __init__(self, terms_path: str = "terms", id_field: str = "id", name_field: str = "name"):
self.terms_path = terms_path
self.id_field = id_field
self.name_field = name_field
def can_handle(self, source_type: str) -> bool:
return source_type.lower() in ["json", "json-ld"]
def parse(self, content: str) -> Dict[str, OntologyTerm]:
"""Parse JSON content."""
terms = {}
data = json.loads(content)
# Navigate to terms array
items = data
if self.terms_path:
for key in self.terms_path.split("."):
if isinstance(items, dict):
items = items.get(key, [])
else:
break
if not isinstance(items, list):
items = [items] if isinstance(items, dict) else []
for item in items:
term = self._item_to_term(item)
if term:
terms[term.id] = term
logger.info(f"Parsed {len(terms)} terms from JSON content")
return terms
def _item_to_term(self, item: Dict) -> Optional[OntologyTerm]:
"""Convert JSON item to OntologyTerm."""
term_id = item.get(self.id_field, "")
name = item.get(self.name_field, "")
if not term_id or not name:
return None
return OntologyTerm(
id=term_id,
name=name,
definition=item.get("definition", item.get("description", "")),
synonyms=item.get("synonyms", item.get("aliases", [])),
xrefs=item.get("xrefs", {}),
is_a=item.get("is_a", item.get("parents", [])),
)
class DatasetLoader:
"""
Main dataset loader supporting multiple formats and sources.
Usage:
loader = DatasetLoader()
loader.load_dataset(config) # Single dataset
loader.load_all_datasets() # From config
"""
def __init__(self, cache_dir: Optional[str] = None):
self.cache_dir = Path(cache_dir or get_config().cache_dir)
self.cache_dir.mkdir(parents=True, exist_ok=True)
# Register adapters
self.adapters: List[DatasetAdapter] = [
OBOAdapter(),
CSVAdapter(),
JSONAdapter(),
]
# Loaded data
self.datasets: Dict[str, Dict[str, OntologyTerm]] = {}
def register_adapter(self, adapter: DatasetAdapter):
"""Register a custom adapter."""
self.adapters.insert(0, adapter) # Custom adapters take priority
def get_adapter(self, source_type: str) -> Optional[DatasetAdapter]:
"""Get adapter for source type."""
for adapter in self.adapters:
if adapter.can_handle(source_type):
return adapter
return None
def load_dataset(self, config: DatasetConfig) -> Dict[str, OntologyTerm]:
"""Load a single dataset based on configuration."""
logger.info(f"Loading dataset: {config.name}")
# Check cache
if config.cache_enabled:
cached = self._load_from_cache(config)
if cached:
self.datasets[config.name] = cached
return cached
# Get content
content = self._get_content(config)
if not content:
logger.warning(f"No content for dataset: {config.name}")
return {}
# Parse with appropriate adapter
adapter = self.get_adapter(config.source_type)
if not adapter:
logger.error(f"No adapter for source type: {config.source_type}")
return {}
terms = adapter.parse(content)
# Cache results
if config.cache_enabled:
self._save_to_cache(config, terms)
self.datasets[config.name] = terms
return terms
def load_all_datasets(self) -> Dict[str, Dict[str, OntologyTerm]]:
"""Load all datasets from configuration."""
config = get_config()
for dataset_config in config.datasets:
self.load_dataset(dataset_config)
return self.datasets
def _get_content(self, config: DatasetConfig) -> Optional[str]:
"""Get content from URL or file path."""
# Try URL first
if config.source_url:
try:
logger.info(f"Downloading from: {config.source_url}")
req = urllib.request.Request(
config.source_url,
headers={'User-Agent': 'HITL-KG/1.0'}
)
with urllib.request.urlopen(req, timeout=60) as response:
return response.read().decode('utf-8')
except Exception as e:
logger.warning(f"Download failed: {e}")
# Try local file
if config.source_path:
path = Path(config.source_path)
if path.exists():
return path.read_text(encoding='utf-8')
return None
def _cache_path(self, config: DatasetConfig) -> Path:
"""Get cache file path for a dataset."""
return self.cache_dir / f"{config.name}_cache.json"
def _load_from_cache(self, config: DatasetConfig) -> Optional[Dict[str, OntologyTerm]]:
"""Load dataset from cache if valid."""
cache_path = self._cache_path(config)
if not cache_path.exists():
return None
# Check age
mtime = datetime.fromtimestamp(cache_path.stat().st_mtime)
age_days = (datetime.now() - mtime).days
if age_days > config.cache_max_age_days:
return None
try:
with open(cache_path) as f:
data = json.load(f)
terms = {}
for term_id, term_data in data.get("terms", {}).items():
terms[term_id] = OntologyTerm(
id=term_data["id"],
name=term_data["name"],
definition=term_data.get("definition", ""),
synonyms=term_data.get("synonyms", []),
xrefs=term_data.get("xrefs", {}),
is_a=term_data.get("is_a", []),
relationships=term_data.get("relationships", []),
namespace=term_data.get("namespace", ""),
)
logger.info(f"Loaded {len(terms)} terms from cache: {config.name}")
return terms
except Exception as e:
logger.warning(f"Cache load failed: {e}")
return None
def _save_to_cache(self, config: DatasetConfig, terms: Dict[str, OntologyTerm]):
"""Save dataset to cache."""
try:
cache_path = self._cache_path(config)
data = {
"name": config.name,
"source_type": config.source_type,
"timestamp": datetime.now().isoformat(),
"terms": {
tid: {
"id": t.id,
"name": t.name,
"definition": t.definition,
"synonyms": t.synonyms,
"xrefs": t.xrefs,
"is_a": t.is_a,
"relationships": t.relationships,
"namespace": t.namespace,
}
for tid, t in terms.items()
}
}
with open(cache_path, 'w') as f:
json.dump(data, f)
logger.info(f"Cached {len(terms)} terms for: {config.name}")
except Exception as e:
logger.warning(f"Cache save failed: {e}")
def build_knowledge_graph(loader: DatasetLoader) -> KnowledgeGraph:
"""
Build a KnowledgeGraph from loaded datasets.
This function:
1. Converts OntologyTerms to Entities
2. Creates relationships between entities
3. Indexes entities for semantic search
"""
kg = KnowledgeGraph()
config = get_config()
# Map dataset names to categories
category_map = {
ds.name: EntityCategory(ds.entity_category)
for ds in config.datasets
if ds.entity_category in [c.value for c in EntityCategory]
}
# Add entities from each dataset
for dataset_name, terms in loader.datasets.items():
category = category_map.get(dataset_name, EntityCategory.FINDING)
for term_id, term in terms.items():
entity = term.to_entity(category)
kg.add_entity(entity)
# Build relationships based on ontology structure
_build_relationships(kg, loader)
logger.info(f"Built KG with {len(kg.entities)} entities")
return kg
def _build_relationships(kg: KnowledgeGraph, loader: DatasetLoader):
"""Build relationships between entities."""
# Disease-symptom associations (curated mappings)
disease_symptom_mappings = _get_disease_symptom_mappings()
for disease_id, symptom_mappings in disease_symptom_mappings.items():
if disease_id not in kg.entities:
continue
for symptom_name, confidence in symptom_mappings:
# Find symptom entity by name
symptom_entity = None
for entity in kg.entities.values():
if entity.category == EntityCategory.SYMPTOM:
if (entity.name.lower() == symptom_name.lower() or
symptom_name.lower() in [s.lower() for s in entity.synonyms]):
symptom_entity = entity
break
if symptom_entity:
kg.add_relation(disease_id, symptom_entity.id, "causes", confidence)
# Add treatment relations
_add_treatment_entities(kg)
def _get_disease_symptom_mappings() -> Dict[str, List[Tuple[str, float]]]:
"""
Get curated disease-symptom mappings.
These are based on medical literature and provide high-quality
associations that may not be present in the raw ontologies.
"""
return {
"DOID:8469": [ # Influenza
("fever", 0.95), ("cough", 0.85), ("fatigue", 0.90),
("body aches", 0.85), ("headache", 0.80), ("chills", 0.75),
],
"DOID:0080600": [ # COVID-19
("fever", 0.80), ("cough", 0.85), ("fatigue", 0.90),
("shortness of breath", 0.70), ("headache", 0.60),
("loss of taste", 0.50), ("loss of smell", 0.50),
],
"DOID:10459": [ # Common cold
("runny nose", 0.95), ("sore throat", 0.80), ("cough", 0.75),
("nasal congestion", 0.85), ("sneezing", 0.80),
],
"DOID:552": [ # Pneumonia
("fever", 0.90), ("cough", 0.95), ("shortness of breath", 0.85),
("chest pain", 0.70), ("fatigue", 0.80),
],
"DOID:6132": [ # Bronchitis
("cough", 0.95), ("fatigue", 0.60),
("shortness of breath", 0.50),
],
"DOID:10534": [ # Strep throat
("sore throat", 0.98), ("fever", 0.80), ("headache", 0.50),
],
"DOID:13084": [ # Sinusitis
("headache", 0.85), ("nasal congestion", 0.90),
("runny nose", 0.80),
],
"DOID:8893": [ # Migraine
("headache", 0.99), ("nausea", 0.70),
],
}
def _add_treatment_entities(kg: KnowledgeGraph):
"""Add treatment entities and relationships."""
treatments = [
Entity("tx_rest", "Rest", EntityCategory.TREATMENT,
"Physical and mental rest", ["bed rest"]),
Entity("tx_fluids", "Fluid Intake", EntityCategory.TREATMENT,
"Increased hydration", ["hydration"]),
Entity("tx_acetaminophen", "Acetaminophen", EntityCategory.MEDICATION,
"Pain and fever reducer", ["paracetamol", "Tylenol"]),
Entity("tx_ibuprofen", "Ibuprofen", EntityCategory.MEDICATION,
"NSAID for pain and inflammation", ["Advil", "Motrin"]),
Entity("tx_antiviral", "Antiviral Medication", EntityCategory.MEDICATION,
"Medications for viral infections", ["oseltamivir", "Tamiflu"]),
Entity("tx_decongestant", "Decongestants", EntityCategory.MEDICATION,
"Nasal congestion relief", ["pseudoephedrine"]),
]
for tx in treatments:
kg.add_entity(tx)
# Treatment relationships
treatment_map = {
"DOID:8469": ["tx_rest", "tx_fluids", "tx_acetaminophen", "tx_antiviral"],
"DOID:0080600": ["tx_rest", "tx_fluids", "tx_acetaminophen"],
"DOID:10459": ["tx_rest", "tx_fluids", "tx_decongestant"],
"DOID:552": ["tx_rest"],
}
for disease_id, treatment_ids in treatment_map.items():
if disease_id in kg.entities:
for tx_id in treatment_ids:
if tx_id in kg.entities:
kg.add_relation(tx_id, disease_id, "treats", 0.8)
def load_knowledge_graph(use_embeddings: bool = True) -> KnowledgeGraph:
"""
Main entry point: Load datasets and build knowledge graph.
Args:
use_embeddings: If True, also index entities for semantic search
"""
loader = DatasetLoader()
loader.load_all_datasets()
kg = build_knowledge_graph(loader)
if use_embeddings:
try:
from .embedding_service import get_embedding_service
embedding_service = get_embedding_service()
embedding_service.index_entities(kg.get_entity_dict_for_embedding())
except Exception as e:
logger.warning(f"Failed to initialize embeddings: {e}")
return kg