This repository contains a cleaned version of the code for ParChain: A Framework for Parallel Hierarchical Agglomerative Clustering using Nearest-Neighbor Chain.

To get the submodules:

git pull
git submodule update --init

Compiler:

• g++ = 7.5.0
• Hardware support for __sync_bool_compare_and_swap_16 needed.

g++ -O3 -std=c++20 -mcx16  -ldl -pthread -I../external/parlaylib/include linkage.cpp -o linkage

run the command in hac folder to compile the version that uses cache tables and range queries.

run the command in hac-matrix folder to compile the version that uses a distance matrix.

./linkage -method [METHOD] -cachesize [cache size] -d [dim] -o [output] [dataset]

• METHOD can be "complete", "ward", "avg" (average linkage with Euclidean distance metric), or "avgsq" (average linkage with squared Euclidean distance metric).
• cache size is the size of each hash table. if cache size=1, no cache will be used.
• output is the output file of the dendrogram

example: ./linkage -method complete -d 2 /path/to/dataset