Skip to content

ANN (Approximate Nearest Neighbor) Indexes

You can create an index over your vector data to make search faster. Vector indexes are faster but less accurate than exhaustive search (KNN or Flat Search). LanceDB provides many parameters to fine-tune the index's size, the speed of queries, and the accuracy of results.

Currently, LanceDB does not automatically create the ANN index. LanceDB has optimized code for KNN as well. For many use-cases, datasets under 100K vectors won't require index creation at all. If you can live with < 100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.

In the future we will look to automatically create and configure the ANN index.

Types of Index

Lance can support multiple index types, the most widely used one is IVF_PQ.

  • IVF_PQ: use Inverted File Index (IVF) to first divide the dataset into N partitions, and then use Product Quantization to compress vectors in each partition.
  • DISKANN (Experimental): organize the vector as a on-disk graph, where the vertices approximately represent the nearest neighbors of each vector.

Creating an IVF_PQ Index

Lance supports IVF_PQ index type by default.

Creating indexes is done via the create_index method.

import lancedb
import numpy as np
uri = "data/sample-lancedb"
db = lancedb.connect(uri)

# Create 10,000 sample vectors
data = [{"vector": row, "item": f"item {i}"}
   for i, row in enumerate(np.random.random((10_000, 1536)).astype('float32'))]

# Add the vectors to a table
tbl = db.create_table("my_vectors", data=data)

# Create and train the index - you need to have enough data in the table for an effective training step
tbl.create_index(num_partitions=256, num_sub_vectors=96)

const vectordb = require('vectordb')
const db = await vectordb.connect('data/sample-lancedb')

let data = []
for (let i = 0; i < 10_000; i++) {
    data.push({vector: Array(1536).fill(i), id: `${i}`, content: "", longId: `${i}`},)
}
const table = await db.createTable('my_vectors', data)
await table.createIndex({ type: 'ivf_pq', column: 'vector', num_partitions: 256, num_sub_vectors: 96 })
  • metric (default: "L2"): The distance metric to use. By default it uses euclidean distance "L2". We also support "cosine" and "dot" distance as well.
  • num_partitions (default: 256): The number of partitions of the index.
  • num_sub_vectors (default: 96): The number of sub-vectors (M) that will be created during Product Quantization (PQ). For D dimensional vector, it will be divided into M of D/M sub-vectors, each of which is presented by a single PQ code.

IVF PQ

IVF_PQ index with num_partitions=2, num_sub_vectors=4

Use GPU to build vector index

Lance Python SDK has experimental GPU support for creating IVF index. Using GPU for index creation requires PyTorch>2.0 being installed.

You can specify the GPU device to train IVF partitions via

  • accelerator: Specify to cuda or mps (on Apple Silicon) to enable GPU training.

# Create index using CUDA on Nvidia GPUs.
tbl.create_index(
     num_partitions=256,
     num_sub_vectors=96,
     accelerator="cuda"
)


# Create index using MPS on Apple Silicon.
tbl.create_index(
     num_partitions=256,
     num_sub_vectors=96,
     accelerator="mps"
)

Trouble shootings:

If you see AssertionError: Torch not compiled with CUDA enabled, you need to install PyTorch with CUDA support.

Querying an ANN Index

Querying vector indexes is done via the search function.

There are a couple of parameters that can be used to fine-tune the search:

  • limit (default: 10): The amount of results that will be returned
  • nprobes (default: 20): The number of probes used. A higher number makes search more accurate but also slower.
    Most of the time, setting nprobes to cover 5-10% of the dataset should achieve high recall with low latency.
    e.g., for 1M vectors divided up into 256 partitions, nprobes should be set to ~20-40.
    Note: nprobes is only applicable if an ANN index is present. If specified on a table without an ANN index, it is ignored.
  • refine_factor (default: None): Refine the results by reading extra elements and re-ranking them in memory.
    A higher number makes search more accurate but also slower. If you find the recall is less than ideal, try refine_factor=10 to start.
    e.g., for 1M vectors divided into 256 partitions, if you're looking for top 20, then refine_factor=200 reranks the whole partition.
    Note: refine_factor is only applicable if an ANN index is present. If specified on a table without an ANN index, it is ignored.

tbl.search(np.random.random((1536))) \
    .limit(2) \
    .nprobes(20) \
    .refine_factor(10) \
    .to_pandas()

                                         vector       item       _distance
0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867


const results_1 = await table
    .search(Array(1536).fill(1.2))
    .limit(2)
    .nprobes(20)
    .refineFactor(10)
    .execute()

The search will return the data requested in addition to the distance of each item.

Filtering (where clause)

You can further filter the elements returned by a search using a where clause.

tbl.search(np.random.random((1536))).where("item != 'item 1141'").to_pandas()

const results_2 = await table
    .search(Array(1536).fill(1.2))
    .where("id != '1141'")
    .execute()

Projections (select clause)

You can select the columns returned by the query using a select clause.

tbl.search(np.random.random((1536))).select(["vector"]).to_pandas()

   vector                                             _distance
0  [0.30928212, 0.022668175, 0.1756372, 0.4911822...  93.971092
1  [0.2525465, 0.01723831, 0.261568, 0.002007689,...  95.173485
...


const results_3 = await table
    .search(Array(1536).fill(1.2))
    .select(["id"])
    .execute()

FAQ

When is it necessary to create an ANN vector index?

LanceDB has manually-tuned SIMD code for computing vector distances. In our benchmarks, computing 100K pairs of 1K dimension vectors takes less than 20ms. For small datasets (< 100K rows) or applications that can accept 100ms latency, vector indices are usually not necessary.

For large-scale or higher dimension vectors, it is beneficial to create vector index.

How big is my index, and how many memory will it take?

In LanceDB, all vector indices are disk-based, meaning that when responding to a vector query, only the relevant pages from the index file are loaded from disk and cached in memory. Additionally, each sub-vector is usually encoded into 1 byte PQ code.

For example, with a 1024-dimension dataset, if we choose num_sub_vectors=64, each sub-vector has 1024 / 64 = 16 float32 numbers. Product quantization can lead to approximately 16 * sizeof(float32) / 1 = 64 times of space reduction.

How to choose num_partitions and num_sub_vectors for IVF_PQ index?

num_partitions is used to decide how many partitions the first level IVF index uses. Higher number of partitions could lead to more efficient I/O during queries and better accuracy, but it takes much more time to train. On SIFT-1M dataset, our benchmark shows that keeping each partition 1K-4K rows lead to a good latency / recall.

num_sub_vectors specifies how many Product Quantization (PQ) short codes to generate on each vector. Because PQ is a lossy compression of the original vector, a higher num_sub_vectors usually results in less space distortion, and thus yields better accuracy. However, a higher num_sub_vectors also causes heavier I/O and more PQ computation, and thus, higher latency. dimension / num_sub_vectors should be a multiple of 8 for optimum SIMD efficiency.