Pinecone Vector Examples

Pinecone Vector Examples

Practical examples for HeliosDB’s Pinecone vector protocol support.

Connection Examples

Python Client

from pinecone import Pinecone

# Connect to HeliosDB
pc = Pinecone(
    api_key="your-api-key",
    host="http://localhost:8080"
)

# Access index
index = pc.Index("my-vectors")

REST API

# Using curl
curl -X POST "http://localhost:8080/vectors/v1/query" \
  -H "Api-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "vector": [0.1, 0.2, 0.3, ...],
    "topK": 10
  }'

Node.js Client

const { Pinecone } = require('@pinecone-database/pinecone');

const pc = new Pinecone({
  apiKey: 'your-api-key',
  environment: 'localhost:8080'
});

const index = pc.index('my-vectors');

Upsert Operations

Single Vector

# Upsert single vector
index.upsert(vectors=[
    {
        "id": "vec1",
        "values": [0.1, 0.2, 0.3, 0.4, 0.5],
        "metadata": {
            "category": "electronics",
            "price": 99.99,
            "in_stock": True
        }
    }
])

Batch Upsert

# Upsert multiple vectors
vectors = [
    {
        "id": f"vec_{i}",
        "values": [random.random() for _ in range(1536)],
        "metadata": {"category": f"cat_{i % 10}"}
    }
    for i in range(1000)
]

# Upsert in batches
index.upsert(vectors=vectors, batch_size=100)

With Namespace

# Upsert to specific namespace
index.upsert(
    vectors=[
        {"id": "prod_1", "values": [...], "metadata": {...}}
    ],
    namespace="production"
)

# Upsert to different namespace
index.upsert(
    vectors=[
        {"id": "test_1", "values": [...], "metadata": {...}}
    ],
    namespace="testing"
)

Query Operations

Basic Query

# Simple similarity search
results = index.query(
    vector=[0.1, 0.2, 0.3, ...],
    top_k=10
)

for match in results.matches:
    print(f"ID: {match.id}, Score: {match.score}")

Query with Metadata

# Include metadata in results
results = index.query(
    vector=[0.1, 0.2, 0.3, ...],
    top_k=10,
    include_metadata=True,
    include_values=True
)

for match in results.matches:
    print(f"ID: {match.id}")
    print(f"Score: {match.score}")
    print(f"Metadata: {match.metadata}")
    print(f"Values: {match.values[:5]}...")

Query with Filter

# Filter by metadata
results = index.query(
    vector=[0.1, 0.2, 0.3, ...],
    top_k=10,
    filter={
        "category": {"$eq": "electronics"},
        "price": {"$lt": 100}
    },
    include_metadata=True
)

Complex Filters

# AND filter
results = index.query(
    vector=[...],
    filter={
        "$and": [
            {"category": {"$eq": "electronics"}},
            {"price": {"$gte": 50}},
            {"in_stock": {"$eq": True}}
        ]
    }
)

# OR filter
results = index.query(
    vector=[...],
    filter={
        "$or": [
            {"category": {"$eq": "electronics"}},
            {"category": {"$eq": "computers"}}
        ]
    }
)

# IN filter
results = index.query(
    vector=[...],
    filter={
        "category": {"$in": ["electronics", "computers", "phones"]}
    }
)

Fetch Operations

Fetch by ID

# Fetch single vector
result = index.fetch(ids=["vec1"])
print(result.vectors["vec1"])

Fetch Multiple

# Fetch multiple vectors
result = index.fetch(ids=["vec1", "vec2", "vec3"])

for id, vector in result.vectors.items():
    print(f"ID: {id}, Values: {vector.values[:5]}...")

Fetch with Namespace

# Fetch from specific namespace
result = index.fetch(
    ids=["prod_1", "prod_2"],
    namespace="production"
)

Delete Operations

Delete by ID

# Delete single vector
index.delete(ids=["vec1"])

# Delete multiple vectors
index.delete(ids=["vec1", "vec2", "vec3"])

Delete by Filter

# Delete by metadata filter
index.delete(
    filter={"category": {"$eq": "outdated"}}
)

Delete All in Namespace

# Delete entire namespace
index.delete(delete_all=True, namespace="testing")

Update Operations

Update Metadata

# Update vector metadata
index.update(
    id="vec1",
    set_metadata={
        "price": 89.99,
        "on_sale": True
    }
)

Update Values

# Update vector values
index.update(
    id="vec1",
    values=[0.2, 0.3, 0.4, ...]  # New vector values
)

Hybrid Search (Sparse + Dense)

Upsert Hybrid Vectors

# Upsert with sparse and dense values
index.upsert(vectors=[
    {
        "id": "doc1",
        "values": [0.1, 0.2, ...],  # Dense embedding
        "sparse_values": {
            "indices": [15, 42, 156, 283],  # Token indices
            "values": [0.8, 0.5, 0.3, 0.2]  # Token weights
        },
        "metadata": {"title": "Introduction to ML"}
    }
])

Hybrid Query

# Query with both dense and sparse vectors
results = index.query(
    vector=[0.1, 0.2, ...],  # Dense query vector
    sparse_vector={
        "indices": [15, 42],
        "values": [0.9, 0.5]
    },
    top_k=10,
    include_metadata=True
)

Index Statistics

Get Index Stats

# Get index statistics
stats = index.describe_index_stats()

print(f"Total vectors: {stats.total_vector_count}")
print(f"Dimension: {stats.dimension}")
print(f"Index fullness: {stats.index_fullness}")

# Per-namespace stats
for ns, ns_stats in stats.namespaces.items():
    print(f"Namespace '{ns}': {ns_stats.vector_count} vectors")

Real-World Use Cases

Semantic Search

from sentence_transformers import SentenceTransformer

# Initialize embedding model
model = SentenceTransformer('all-MiniLM-L6-v2')

# Index documents
documents = [
    {"id": "doc1", "text": "Machine learning fundamentals"},
    {"id": "doc2", "text": "Deep learning architectures"},
    {"id": "doc3", "text": "Natural language processing"}
]

vectors = [
    {
        "id": doc["id"],
        "values": model.encode(doc["text"]).tolist(),
        "metadata": {"text": doc["text"]}
    }
    for doc in documents
]

index.upsert(vectors=vectors)

# Search
query = "neural networks"
query_embedding = model.encode(query).tolist()

results = index.query(
    vector=query_embedding,
    top_k=5,
    include_metadata=True
)

for match in results.matches:
    print(f"Score: {match.score:.4f} - {match.metadata['text']}")

RAG (Retrieval-Augmented Generation)

import openai

def rag_query(question, index, model, top_k=5):
    # Embed question
    question_embedding = model.encode(question).tolist()

    # Retrieve relevant documents
    results = index.query(
        vector=question_embedding,
        top_k=top_k,
        include_metadata=True
    )

    # Build context from retrieved documents
    context = "\n".join([
        m.metadata.get("text", "")
        for m in results.matches
    ])

    # Generate answer with LLM
    response = openai.ChatCompletion.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": f"Context:\n{context}"},
            {"role": "user", "content": question}
        ]
    )

    return response.choices[0].message.content

# Use RAG
answer = rag_query(
    "What is deep learning?",
    index,
    model
)
print(answer)

Image Similarity Search

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

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

# Remove classification layer
model = torch.nn.Sequential(*list(model.children())[:-1])

# Transform for images
transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

def get_image_embedding(image_path):
    img = Image.open(image_path).convert('RGB')
    img_tensor = transform(img).unsqueeze(0)
    with torch.no_grad():
        embedding = model(img_tensor)
    return embedding.squeeze().numpy().tolist()

# Index images
for image_path in image_paths:
    embedding = get_image_embedding(image_path)
    index.upsert(vectors=[{
        "id": image_path,
        "values": embedding,
        "metadata": {"path": image_path}
    }])

# Search similar images
query_embedding = get_image_embedding("query_image.jpg")
results = index.query(
    vector=query_embedding,
    top_k=10,
    include_metadata=True
)

Recommendation System

# Index user preferences
user_vectors = [
    {
        "id": f"user_{user_id}",
        "values": user_embedding,
        "metadata": {
            "user_id": user_id,
            "preferences": preferences
        }
    }
    for user_id, user_embedding, preferences in users
]

index.upsert(vectors=user_vectors, namespace="users")

# Index items
item_vectors = [
    {
        "id": f"item_{item_id}",
        "values": item_embedding,
        "metadata": {
            "item_id": item_id,
            "category": category,
            "price": price
        }
    }
    for item_id, item_embedding, category, price in items
]

index.upsert(vectors=item_vectors, namespace="items")

# Get recommendations for user
def recommend(user_id, top_k=10):
    # Get user vector
    user = index.fetch(
        ids=[f"user_{user_id}"],
        namespace="users"
    )
    user_vector = user.vectors[f"user_{user_id}"].values

    # Find similar items
    results = index.query(
        vector=user_vector,
        top_k=top_k,
        include_metadata=True,
        namespace="items"
    )

    return [
        {
            "item_id": m.metadata["item_id"],
            "category": m.metadata["category"],
            "score": m.score
        }
        for m in results.matches
    ]

recommendations = recommend("user_123")

Anomaly Detection

def detect_anomalies(vector, threshold=0.8):
    """
    Detect if a vector is anomalous by comparing
    to existing vectors in the index.
    """
    results = index.query(
        vector=vector,
        top_k=5
    )

    # If no similar vectors found, it's anomalous
    if not results.matches:
        return True, 0.0

    # Average similarity score
    avg_score = sum(m.score for m in results.matches) / len(results.matches)

    # If average similarity is below threshold, it's anomalous
    is_anomaly = avg_score < threshold

    return is_anomaly, avg_score

# Check new data point
new_vector = [0.1, 0.2, ...]
is_anomaly, score = detect_anomalies(new_vector)

if is_anomaly:
    print(f"Anomaly detected! Similarity score: {score}")

---

Related: README.md | CONFIGURATION.md | COMPATIBILITY.md

Last Updated: December 2025