from sentence_transformers import SentenceTransformer
model = SentenceTransformer('all-MiniLM-L6-v2')
embedding = model.encode(text)

import feather_db
from sentence_transformers import SentenceTransformer

model = SentenceTransformer('all-MiniLM-L6-v2')
db = feather_db.DB.open("docs.feather", dim=384)

# Add documents
for i, doc in enumerate(documents):
    embedding = model.encode(doc)
    db.add(i, embedding)

# Search
query_emb = model.encode("your query")
ids, distances = db.search(query_emb, k=5)

import feather_db
from torchvision import models, transforms
from PIL import Image

# Load image model
model = models.resnet50(pretrained=True)
model.eval()

db = feather_db.DB.open("images.feather", dim=2048)

# Add images
for i, img_path in enumerate(image_paths):
    img = Image.open(img_path)
    embedding = extract_features(model, img)
    db.add(i, embedding)

# Find similar images
query_emb = extract_features(model, query_image)
ids, distances = db.search(query_emb, k=10)

import feather_db
import numpy as np

db = feather_db.DB.open("products.feather", dim=256)

# Add product embeddings
for product_id, features in products.items():
    embedding = create_product_embedding(features)
    db.add(product_id, embedding)

# Find similar products
product_emb = get_product_embedding(product_id)
similar_ids, _ = db.search(product_emb, k=5)

from sentence_transformers import SentenceTransformer
model = SentenceTransformer('all-MiniLM-L6-v2')  # 384 dims
# or
model = SentenceTransformer('all-mpnet-base-v2')  # 768 dims

from torchvision import models
model = models.resnet50(pretrained=True)

# Good: Process in batches
for i in range(0, len(data), 1000):
    batch = data[i:i+1000]
    for item in batch:
        db.add(item.id, item.embedding)
    db.save()  # Save periodically

# Bad: Save after every add
for item in data:
    db.add(item.id, item.embedding)
    db.save()  # Too slow!

try:
    db = feather_py.DB.open("db.feather", dim=768)
    db.add(id=1, vec=vector)
except RuntimeError as e:
    print(f"Error: {e}")
finally:
    db.save()

def normalize(vec):
    norm = np.linalg.norm(vec)
    return vec / norm if norm > 0 else vec

# Use for cosine similarity-like behavior
normalized = normalize(embedding)
db.add(id=1, vec=normalized)