Crate ecp_core 

The core library of ecpfs. It builds and searches Extended Cluster Pruning (eCP) indexes, stored as zarr arrays on disk.

• build::builder::Builder builds an index from embeddings in an .h5 or .zarr file.
• search::Index opens an index for k-nearest-neighbor search and inserts.
• search::IndexInfo reads an index’s settings without loading its tree.
• logging::init turns on logging to a JSONL file.
• error::EcpError is what a fallible call in this crate returns on failure.
• format::FORMAT_VERSION is the on-disk format version this build writes and reads.

The ecpfs Python package and the ecp command line tool both wrap this crate.

Re-exports

pub use error::EcpError;

pub use error::Result;

Modules

build

Building an index, from picking representatives to writing the tree to disk.

dtype

The types embeddings can be stored as on disk, and reading them back as f32.

error

ecp-core’s error type. Every fallible function in this crate returns Result, std::result::Result<T, EcpError>.

format

The on-disk index format’s version, and the arrays a version 1 index holds.

logging

Optional logging to a JSONL file, one JSON object per line.

metric

Distance metrics and the distance computations built on them.

search

Opening a built index to search it and insert into it.

utils

The default memory limit. Also re-exports the metric and dtype items.