Detailed Experimental Results for ALENEX'17 Submission "Engineering a direct k-way Hypergraph Partitioning Algorithm"

This website accompanies our ALENEX'17 submission. We provide interactive visualizations of the performance plots as well as detailed per-instance results in .csv format.

Comparison of KaHyPar-K and KaHyPar-R

The performance plots comparing KaHyPar-K with KaHyPar-R and KaHyPar-R* can be viewed here for hypergraphs with $|\tilde{e}| \geq 3$ and here for hypergraphs with $|\tilde{e}| < 3$ .

Comparison of KaHyPar-K and other Systems

The performance plots comparing KaHyPar-K with hMetis-R, hMetis-K, PaToH-Q, and PaToH-D can be viewed here for hypergraphs with $|\tilde{e}| \geq 3$ and here for hypergraphs with $|\tilde{e}| < 3$ .
Additionally this plot shows the partitioning results for all hypergraphs with $|\tilde{e}| \geq 28$ .

Comparison of KaHyPar-K and other Systems for non-powers of two

Detailed Per-Instance Results

We provide detailed per-instance results for each hypergraph partitioner. For each instance, the .csv files contain the following information:

hypergraph name
median hyperedge size
k
minimum cut (connectivity metric): min_km1
average cut (connectivity metric): avg_km1
average imbalance
average running time

The .csv files for the full benchmark set can be found here:

The .csv files for non-powers of two can be found here (Note that these files do not contain average running times, because experiments were performed in parallel on a single machine.):

The list of excluded instances is available here.

Hypergraph Statistics

Detailed statistics of the hypergraphs used in the experimental evaluation can be found here in .csv format.