ACTIVE

ACTIVE: A Dynamic Pareto-Frontier-Based Unified Index for Dual-Vector Query. Supports efficient build, search, and streaming update operations. The full version of the paper is available here.

Prerequisites

• C++17 or later
• CMake >= 2.8
• Boost (for dynamic_bitset)
• TBB (Intel Threading Building Blocks)
• OpenMP

Install dependencies on Ubuntu/Debian:

sudo apt install build-essential cmake libboost-all-dev libtbb-dev

Compilation

cd ACTIVE
mkdir -p build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j$(nproc)

The compiled executables will be located under build/test/.

To compile individual tools under tools/ (data preparation utilities):

mkdir -p tools/exe
g++ tools/gen_alpha.cpp -o tools/exe/gen_alpha
g++ tools/process_base.cpp -o tools/exe/process_base
g++ tools/gen_trace.cpp -o tools/exe/gen_trace
g++ tools/gen_gt.cpp -o tools/exe/gen_gt -fopenmp

Four Functionalities

1. Build Index

Build a DEG (Distance-Enhanced Graph) index from base embedding and location vectors.

Item	Path
Source	test/build_index.cpp
Executable	build/test/build_index

Usage:

./build_index <vec_base_emb> <vec_base_loc> <graph_index> <ef(L)> <max_edges(R)> <threads> <max_emb_dis> <max_loc_dis>

2. Search

Search the index for k-nearest neighbors given a query set with alpha-weighted distance.

Item	Path
Source	test/search.cpp
Executable	build/test/search

Usage:

./search <vec_base_emb> <vec_base_loc> <graph_index> <L> <K> <alpha_file> <query_emb> <query_loc> <query_gt> <threads> <max_emb_dis> <max_loc_dis> <search_type>

3. Insert

Insert new data points into an existing DEG index.

Item	Path
Source	test/insert.cpp
Executable	build/test/insert

Usage:

./insert <graph> <base_emb> <base_loc> <trace1> [trace2 ...] <gt1> [gt2 ...] <ef> <max_edges> <threads> <id_flag> <delete_mode> <L> <K> <angle> <query_alpha> <query_emb> <query_loc>

4. Update (Single-Round)

Perform a single-round update (insert / delete / both) on the index, with search evaluation between rounds.

Item	Path
Source	test/update.cpp
Executable	build/test/update

Usage:

./update <graph_path> <base_emb> <base_loc> <trace_path> <ef(L)> <max_edges(R)> <threads> <id_flag> [L K query_alpha query_emb query_loc query_gt] <delete_mode>

---

Data Preparation Scripts

Scripts for generating datasets, traces, and ground truth:

Script	Description
tools/process_dataset.sh	Full pipeline: gen_alpha, process_base, gen_trace, gen_gt
tools/gen_trace_gt.sh	Generate update traces and ground truth for streaming benchmarks
tools/et_base.sh	Dataset preprocessing with sliding-window ground truth generation

Data Format

The index works with .fvecs (float vectors) and .ivecs (integer vectors) binary formats:

• Each file starts with a 4-byte dimension header
• Each vector is stored as (dim, values...) where dim is a 4-byte integer followed by dim elements

Project Structure

├── include/          # Header files (index, component, builder, etc.)
├── src/              # Library source files
├── test/             # Executable entry points (build, search, insert, update)
├── tools/            # Data preparation utilities
├── CMakeLists.txt    # Top-level build configuration
└── ACTIVE-FULL.pdf   # Full paper