mbishu's picture
Initialize SDK and fix vector store issues
a475b1f
Raw History Blame Contribute Delete
15.9 kB
"""
Pure Python SDK Client for Image Search.
This SDK uses the actual services from app/services/ directly,
providing the same functionality as the API but as a Python library.
Usage:
from image_search_sdk import ImageSearchClient
# Initialize with your vector store choice
client = ImageSearchClient(
data_dir="./my_data",
model_name="clip-ViT-L-14",
vector_store_type="pinecone", # or qdrant, weaviate, chromadb, pgvector
vector_store_config={
"api_key": "your_pinecone_key",
"cloud": "aws",
"region": "us-east-1"
}
)
# Connect your product database
client.connect_database(
connection_string="sqlite:///products.db",
id_column="product_id",
title_column="name"
)
# First-time: Index all existing products
stats = client.initialize()
# Ongoing: Add new products (embeddings generated automatically)
client.add_product(new_product)
# Search by image
results = client.search_by_image("query.jpg")
"""
import time
from typing import List, Optional, Union, Dict, Any
from pathlib import Path
from PIL import Image
import io
from .models import Product, SearchResult, SearchOptions
from .exceptions import ImageSearchError, ValidationError
from .schema_mapper import SchemaMapper, SchemaMappingBuilder
class ImageSearchClient:
"""
Pure Python SDK client that uses services directly.
No API calls - everything runs locally using the actual services
from app/services/ (CLIP, vector stores, reranking, etc.)
"""
def __init__(
self,
data_dir: str = "./sdk_data",
model_name: str = "clip-ViT-L-14",
vector_store_type: str = "chromadb", # pinecone, qdrant, weaviate, chromadb, pgvector
vector_store_config: Optional[Dict[str, Any]] = None
):
"""
Initialize the Image Search SDK.
Args:
data_dir: Directory for storing data
model_name: CLIP model ("clip-ViT-B-32" or "clip-ViT-L-14")
vector_store_type: Vector store (pinecone, qdrant, weaviate, chromadb, pgvector)
vector_store_config: Vector store connection parameters
Examples:
# Pinecone (Managed, Serverless)
client = ImageSearchClient(
vector_store_type="pinecone",
vector_store_config={"api_key": "xxx", "cloud": "aws", "region": "us-east-1"}
)
# Qdrant (Self-hosted)
client = ImageSearchClient(
vector_store_type="qdrant",
vector_store_config={"mode": "local", "path": "./qdrant_data"}
)
# ChromaDB (Simple, No setup)
client = ImageSearchClient(
vector_store_type="chromadb",
vector_store_config={"mode": "local", "path": "./chroma_data"}
)
"""
import sys
from pathlib import Path as P
# Add app/services to Python path
app_services_path = P(__file__).parent.parent.parent.parent / "app"
if str(app_services_path) not in sys.path:
sys.path.insert(0, str(app_services_path))
# Import unified services
from services.ai.clip_service import ClipEmbeddingService
from services.vector_stores.manager import create_vector_store_manager
from services.database import create_database_adapter as _create_db_adapter, DatabaseConfig as _DatabaseConfig, DatabaseError
# Store for later use
self._create_database_adapter = _create_db_adapter
self._DatabaseConfig = _DatabaseConfig
self.DatabaseError = DatabaseError
self.data_dir = Path(data_dir)
self.data_dir.mkdir(exist_ok=True)
# Initialize CLIP service
self.clip_service = ClipEmbeddingService(model_name=model_name)
# Initialize vector store manager
self.vector_store_manager = create_vector_store_manager(
store_type=vector_store_type,
connection_params=vector_store_config or {},
clip_service=self.clip_service,
collection_name="product_embeddings",
state_file=str(self.data_dir / "init_state.json")
)
self.database_adapter: Optional[DatabaseAdapter] = None
self.schema_mapper: Optional[SchemaMapper] = None
self.product_monitor: Optional[Any] = None # Auto-monitor for new products
def connect_database(
self,
connection_string: str,
table_name: str = "products",
schema_mapping: Optional[Dict[str, Any]] = None,
**column_mappings
) -> None:
"""
Connect to your existing product database with flexible schema mapping.
Args:
connection_string: Database connection string
table_name: Name of your products table
schema_mapping: Complete custom schema mapping dict for complex structures
**column_mappings: Simple column mappings (id_column, title_column, etc.)
Examples:
# Simple mapping
client.connect_database(
"sqlite:///products.db",
id_column="product_id",
title_column="name",
image_url_column="image",
category_column="cat",
price_column="price"
)
# Complex custom mapping
client.connect_database(
"postgresql://user:pass@localhost/db",
schema_mapping={
"table_name": "products",
"id_mapping": {"source_column": "sku", "target_field": "id"},
"title_mapping": {"source_column": "product_name", "target_field": "title"},
...
}
)
"""
# Build schema mapping
if schema_mapping:
schema = SchemaMappingBuilder.from_dict(schema_mapping)
else:
# Create simple mapping from keyword arguments
schema = SchemaMappingBuilder.create_simple_mapping(
table_name=table_name,
id_column=column_mappings.get('id_column', 'id'),
title_column=column_mappings.get('title_column', 'title'),
image_url_column=column_mappings.get('image_url_column', 'image_url'),
category_column=column_mappings.get('category_column'),
brand_column=column_mappings.get('brand_column'),
price_column=column_mappings.get('price_column'),
rating_column=column_mappings.get('rating_column'),
description_column=column_mappings.get('description_column')
)
self.schema_mapper = SchemaMapper(schema)
# Create database adapter
config = self._DatabaseConfig(
connection_string=connection_string,
table_name=table_name,
id_column=column_mappings.get('id_column', 'id'),
title_column=column_mappings.get('title_column', 'title'),
image_url_column=column_mappings.get('image_url_column', 'image_url')
)
self.database_adapter = self._create_database_adapter(config)
self.database_adapter.connect()
def enable_auto_monitor(self, poll_interval: int = 60, enable: bool = True):
"""
Enable automatic monitoring for new products.
When enabled, the system will automatically detect new products
added to your database and generate embeddings for them.
Args:
poll_interval: How often to check for new products (seconds)
enable: True to enable, False to disable
"""
if not self.database_adapter:
raise self.DatabaseError("Connect database first")
if enable:
# Import monitor
from services.database.product_monitor import ProductMonitor
# Create and start monitor
self.product_monitor = ProductMonitor(
database_adapter=self.database_adapter,
vector_store_manager=self.vector_store_manager,
poll_interval=poll_interval,
batch_size=10
)
self.product_monitor.start()
else:
# Stop monitoring
if self.product_monitor:
self.product_monitor.stop()
self.product_monitor = None
def initialize(self, batch_size: int = 50) -> Dict[str, Any]:
"""
First-time initialization: Generate embeddings for ALL existing products.
This should be called once after connecting to your database.
Progress is saved, so you can resume if interrupted.
Args:
batch_size: Number of products to process in each batch
Returns:
Dictionary with indexing statistics
"""
if not self.database_adapter:
raise self.DatabaseError("No database connected. Call connect_database() first.")
start_time = time.time()
# Use vector store manager for initialization
stats = self.vector_store_manager.initialize_from_database(
products_iterator=self.database_adapter.get_all_products(),
batch_size=batch_size,
progress_callback=None # Can add progress callback if needed
)
duration = time.time() - start_time
stats['duration_seconds'] = duration
stats['products_per_second'] = stats['indexed_products'] / duration if duration > 0 else 0
return stats
def add_product(self, product: Product) -> bool:
"""
Add a new product with automatic embedding generation.
Embeddings are generated automatically when you add products.
Args:
product: Product to add
Returns:
True if successful
"""
return self.vector_store_manager.add_product(product)
def search_by_image(
self,
image_path: Union[str, Path],
options: Optional[SearchOptions] = None
) -> List[SearchResult]:
"""
Search for similar products using an image.
Args:
image_path: Path to the query image
options: Search options and filters
Returns:
List of search results
"""
if options is None:
options = SearchOptions()
try:
# Load and encode image
image = Image.open(image_path).convert("RGB")
query_embedding = self.clip_service.encode_image(image)
# Build filters
filters = {}
if options.category:
filters['category'] = options.category
if options.brand:
filters['brand'] = options.brand
# Search using vector store manager
results = self.vector_store_manager.search(
query_embedding=query_embedding,
top_k=options.top_k,
filters=filters if filters else None
)
# Convert to SearchResult objects
search_results = []
for result in results:
product = Product(
id=result['product_id'],
title=result['metadata'].get('title', ''),
image_url='',
category=result['metadata'].get('category'),
brand=result['metadata'].get('brand'),
price=result['metadata'].get('price'),
rating=result['metadata'].get('rating')
)
search_results.append(SearchResult(
product=product,
similarity_score=result['score'],
rank=len(search_results) + 1,
search_time_ms=0
))
return search_results
except Exception as e:
raise ImageSearchError(f"Image search failed: {e}")
def search_by_text(
self,
query: str,
options: Optional[SearchOptions] = None
) -> List[SearchResult]:
"""
Search for products using text query.
Args:
query: Text search query
options: Search options and filters
Returns:
List of search results
"""
if options is None:
options = SearchOptions()
try:
# Generate text embedding
query_embedding = self.clip_service.encode_text(query)
# Build filters
filters = {}
if options.category:
filters['category'] = options.category
if options.brand:
filters['brand'] = options.brand
# Search using vector store manager
results = self.vector_store_manager.search(
query_embedding=query_embedding,
top_k=options.top_k,
filters=filters if filters else None
)
# Convert to SearchResult objects
search_results = []
for result in results:
product = Product(
id=result['product_id'],
title=result['metadata'].get('title', ''),
image_url='',
category=result['metadata'].get('category'),
brand=result['metadata'].get('brand'),
price=result['metadata'].get('price'),
rating=result['metadata'].get('rating')
)
search_results.append(SearchResult(
product=product,
similarity_score=result['score'],
rank=len(search_results) + 1,
search_time_ms=0
))
return search_results
except Exception as e:
raise ImageSearchError(f"Text search failed: {e}")
def update_product(self, product: Product) -> bool:
"""
Update an existing product - regenerates embeddings automatically.
Args:
product: Updated product
Returns:
True if successful
"""
return self.vector_store_manager.update_product(product)
def delete_product(self, product_id: str) -> bool:
"""
Delete a product and its embeddings.
Args:
product_id: ID of product to delete
Returns:
True if successful
"""
return self.vector_store_manager.delete_product(product_id)
def get_stats(self) -> Dict[str, Any]:
"""
Get system statistics.
Returns:
Dictionary with system stats
"""
return self.vector_store_manager.get_stats()
def close(self) -> None:
"""Close all connections and cleanup resources."""
# Stop monitoring
if self.product_monitor:
self.product_monitor.stop()
# Close database
if self.database_adapter:
self.database_adapter.disconnect()
# Close vector store
if self.vector_store_manager:
self.vector_store_manager.close()