Sparse Vector Search Setup

from chromadb import Schema, SparseVectorIndexConfig, K
from chromadb.utils.embedding_functions import ChromaCloudSpladeEmbeddingFunction

schema = Schema()

# Add sparse vector index for keyword-based search
# "sparse_embedding" is just a metadata key name - use any name you prefer
sparse_ef = ChromaCloudSpladeEmbeddingFunction()
schema.create_index(
    config=SparseVectorIndexConfig(
        source_key=K.DOCUMENT,
        embedding_function=sparse_ef
    ),
    key="sparse_embedding"
)

import { Schema, SparseVectorIndexConfig, K } from 'chromadb';
import { ChromaCloudSpladeEmbeddingFunction } from '@chroma-core/chroma-cloud-splade';

const schema = new Schema();

// Add sparse vector index for keyword-based search
// "sparse_embedding" is just a metadata key name - use any name you prefer
const sparseEf = new ChromaCloudSpladeEmbeddingFunction({
  apiKeyEnvVar: "CHROMA_API_KEY"
});
schema.createIndex(
  new SparseVectorIndexConfig({
    sourceKey: K.DOCUMENT,
    embeddingFunction: sparseEf
  }),
  "sparse_embedding"
);

import chromadb

client = chromadb.CloudClient(
    tenant="your-tenant",
    database="your-database",
    api_key="your-api-key"
)

collection = client.create_collection(
    name="hybrid_search_collection",
    schema=schema
)

import { CloudClient } from 'chromadb';

const client = new CloudClient({
  tenant: "your-tenant",
  database: "your-database",
  apiKey: "your-api-key"
});

const collection = await client.createCollection({
  name: "hybrid_search_collection",
  schema: schema
});

collection.add(
    ids=["doc1", "doc2", "doc3"],
    documents=[
        "The quick brown fox jumps over the lazy dog",
        "A fast auburn fox leaps over a sleepy canine",
        "Machine learning is a subset of artificial intelligence"
    ],
    metadatas=[
        {"category": "animals"},
        {"category": "animals"},
        {"category": "technology"}
    ]
)

# Sparse embeddings for "sparse_embedding" are generated automatically
# from the documents (source_key=K.DOCUMENT)

await collection.add({
  ids: ["doc1", "doc2", "doc3"],
  documents: [
    "The quick brown fox jumps over the lazy dog",
    "A fast auburn fox leaps over a sleepy canine",
    "Machine learning is a subset of artificial intelligence"
  ],
  metadatas: [
    { category: "animals" },
    { category: "animals" },
    { category: "technology" }
  ]
});

// Sparse embeddings for "sparse_embedding" are generated automatically
// from the documents (source_key=K.DOCUMENT)

from chromadb import Search, K, Knn

# Search using sparse embeddings only
sparse_rank = Knn(query="fox animal", key="sparse_embedding")

# Build and execute search
search = (Search()
    .rank(sparse_rank)
    .limit(10)
    .select(K.DOCUMENT, K.SCORE))

results = collection.search(search)

# Process results
for row in results.rows()[0]:
    print(f"Score: {row['score']:.3f} - {row['document']}")

import { Search, K, Knn } from 'chromadb';

// Search using sparse embeddings only
const sparseRank = Knn({ query: "fox animal", key: "sparse_embedding" });

// Build and execute search
const search = new Search()
  .rank(sparseRank)
  .limit(10)
  .select(K.DOCUMENT, K.SCORE);

const results = await collection.search(search);

// Process results
for (const row of results.rows()[0]) {
  console.log(`Score: ${row.score.toFixed(3)} - ${row.document}`);
}

from chromadb import Search, K, Knn, Rrf

# Create RRF ranking combining dense and sparse embeddings
hybrid_rank = Rrf(
    ranks=[
        Knn(query="fox animal", return_rank=True),           # Dense semantic search
        Knn(query="fox animal", key="sparse_embedding", return_rank=True)  # Sparse keyword search
    ],
    weights=[0.7, 0.3],  # 70% semantic, 30% keyword
    k=60
)

# Build and execute search
search = (Search()
    .rank(hybrid_rank)
    .limit(10)
    .select(K.DOCUMENT, K.SCORE))

results = collection.search(search)

# Process results
for row in results.rows()[0]:
    print(f"Score: {row['score']:.3f} - {row['document']}")

import { Search, K, Knn, Rrf } from 'chromadb';

// Create RRF ranking combining dense and sparse embeddings
const hybridRank = Rrf({
  ranks: [
    Knn({ query: "fox animal", returnRank: true }),           // Dense semantic search
    Knn({ query: "fox animal", key: "sparse_embedding", returnRank: true })  // Sparse keyword search
  ],
  weights: [0.7, 0.3],  // 70% semantic, 30% keyword
  k: 60
});

// Build and execute search
const search = new Search()
  .rank(hybridRank)
  .limit(10)
  .select(K.DOCUMENT, K.SCORE);

const results = await collection.search(search);

// Process results
for (const row of results.rows()[0]) {
  console.log(`Score: ${row.score.toFixed(3)} - ${row.document}`);
}