IndexKind Enum

Index kinds for similarity search.

HNSW Hierarchical Navigable Small World which performs an approximate nearest neighbor (ANN) search. Lower accuracy than exhaustive k nearest neighbor, but faster and more efficient.

Flat Does a brute force search to find the nearest neighbors. Calculates the distances between all pairs of data points, so has a linear time complexity, that grows directly proportional to the number of points. Also referred to as exhaustive k nearest neighbor in some databases. High recall accuracy, but slower and more expensive than HNSW. Better with smaller datasets.

IVF Flat Inverted File with Flat Compression. Designed to enhance search efficiency by narrowing the search area through the use of neighbor partitions or clusters. Also referred to as approximate nearest neighbor (ANN) search.

Disk ANN Disk-based Approximate Nearest Neighbor algorithm designed for efficiently searching for approximate nearest neighbors (ANN) in high-dimensional spaces. The primary focus of DiskANN is to handle large-scale datasets that cannot fit entirely into memory, leveraging disk storage to store the data while maintaining fast search times.

Quantized Flat Index that compresses vectors using DiskANN-based quantization methods for better efficiency in the kNN search.

Dynamic Dynamic index allows to automatically switch from FLAT to HNSW indexes.

Fields

DISK_ANN

DYNAMIC

FLAT

HNSW

IVF_FLAT

QUANTIZED_FLAT