hnswlab/README.md

AI Data Index Design

1. Testing Steps

1. Make sure CMake and other build tools are installed:

   sudo apt-get install cmake build-essentials
   

2. Create a \build folder inside the hnswlab directory.

3. Change directory to the build folder:

   cd build
   

4. Run CMake to generate the build files:

   cmake ..
   

5. Build the project:

   make
   

6. Run the test program:

   ./hnsw_test data_file_path data_size query_file_path groundtruth_file_path
   

   For example:

   ./hnsw_test ../dataset/siftsmall/siftsmall_base.fvecs 10000 ../dataset/siftsmall/siftsmall_query.fvecs 100 ../dataset/siftsmall/siftsmall_groundtruth.ivecs
   

   Our test program will report the recall value and time costs of your algorithm.

2. Mission Description

You need to implement two functions inside hnsw.h and hnsw.c in HNSW way:

HNSWContext *hnsw_init_context(const char *filename, size_t dim, size_t len); // load data and build graph
void hnsw_approximate_knn(HNSWContext *ctx, VecData *q, int *results, int k); // search KNN results

We have implemented data loading and provided a simplest KNN algorithm. But our implementation can only handle small batches of data(SIFTSMALL dataset), please implement a new approximate KNN algorithm based on the HNSW algorithm so that it can handle large batches of data(SIFT dataset) efficiently.

3. Data Download

Please visit http://corpus-texmex.irisa.fr/

TODO: We should provide a script to download datasets automatically