Authors: Gaurav Gupta, Tharun Medini, Anshumali Shrivastava, Alexander J. Smola
Representation learning has transformed the problem of information retrieval into one of finding the approximate set of nearest neighbors in a high dimensional vector space. With limited hardware resources and time-critical queries, the retrieval engines face an inherent tension between latency, accuracy, scalability, compactness, and the ability to load balance in distributed settings. To improve the trade-off, we propose a new algorithm, called BaLanced Index for Scalable Search (BLISS), a highly tunable indexing algorithm with enviably small index sizes, making it easy to scale to billions of vectors. It iteratively refines partitions of items by learning the relevant buckets directly from the query-item relevance data. To ensure that the buckets are balanced, BLISS uses the power-of-𝐾 choices strategy. We show that BLISS provides superior load balancing with high probability (and under very benign assumptions). Due to its design, BLISS can be employed for both near-neighbor retrieval (ANN problem) and extreme classification (XML problem). For the case of ANN, we train and index 4 datasets with billion vectors each. We compare the recall, inference time, indexing time, and index size for BLISS with the two most popular and well-optimized libraries- Hierarchical Navigable Small World (HNSW) graph and Facebook’s FAISS. BLISS requires 100× lesser RAM than HNSW, making it fit in memory on commodity machines while taking a similar inference time as HNSW for the same recall. Against FAISS-IVF, BLISS achieves similar performance with 3-4× less memory requirement. BLISS is both data and model parallel, making it ideal for distributed implementation for training and inference. For the case of XML, BLISS surpasses the best baselines’ precision while being 5× faster for inference on popular multi-label datasets with half a million classes.
Compiler: gcc and g++
pip install -r requirements.txt
cd src/InvertedIndex && make
Create a directory named "data" outside the BLISS repo. Create a sperate directory for each dataset. For example download the GloVe data:
cd data
mkdir glove && cd glove
wget http:https://ann-benchmarks.com/glove-100-angular.hdf5
The chosen dataset folder names should match with the BLISS/src/config.py file.
The GloVe and SIFT datasets came from ann-benchmarks. The Billion scale data came from big-ann-benchmarks. The datasets for extreme classification can be downloaded from the XMLRepository
Update the BLISS/src/config.py DATASET dictionary after adding any new dataset.
- Train the iterative model
python3 construct.py --index='glove_epc20_K2_B4096_R4'
- Index on the trained model
python3 index.py --index='glove_epc20_K2_B4096_R4'
- Query the index
python3 query.py --index='glove_epc20_K2_B4096_R4' --topm=50
We'd love to accept your contributions to this project. Please feel free to open an issue, or submit a pull request as necessary.
The code is build upon Tharun24/MACH
If you find the provided code useful, please cite our work with the following bibtex.
@inproceedings{10.1145/3534678.3539414,
author = {Gupta, Gaurav and Medini, Tharun and Shrivastava, Anshumali and Smola, Alexander J.},
title = {BLISS: A Billion Scale Index Using Iterative Re-Partitioning},
year = {2022},
isbn = {9781450393850},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3534678.3539414},
doi = {10.1145/3534678.3539414},
booktitle = {Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages = {486–495},
numpages = {10},
keywords = {classification, search, load-balance, billion-scale, learning-to-index},
location = {Washington DC, USA},
series = {KDD '22}
}TODO:
- Multithreading support for score aggregation and re-ranking.