All datasets must be in the AnnDataset format. You can find datasets used in our paper in this bucket.
You can run any number of experiments with the invocation of the following command line:
cargo run --release -- --config ${PATH_TO_YAML_CONFIG}
The YAML configuration file must contain a manifest of all experiments
you wish to run. For a detailed structure of the configuration, see
crate::experiment::Config
. You can find configs used in
our paper along with output logs in experiments/
.
The degree of parallelism of the code can be set through the RAYON_NUM_THREADS env variable. Here is an example setting the computation up to 64 threads:
export RAYON_NUM_THREADS=64
Please use the following to cite our work:
@misc{bruch2024optimistic,
title={Optimistic Query Routing in Clustering-based Approximate Maximum Inner Product Search},
author={Sebastian Bruch and Aditya Krishnan and Franco Maria Nardini},
year={2024},
eprint={2405.12207},
archivePrefix={arXiv},
primaryClass={cs.LG}
}