conashin/hgp-benchmark-unweighted

这是一个用于双向超图划分（在平衡约束下的最小割）的转换基准数据集。它源自集成基准集合（位于Zenodo 15373339），并重新格式化为适合EDA划分工具链的普通未加权网表格式。数据集包含489个案例，单元数量范围从5,804到9,128,020，网络数量范围从163到6,920,306，引脚数量范围从6,323到57,156,537，平均度范围从1.45到1,825。原始格式为hMETIS .hgr文件，转换后为未加权网表格式。数据集按引脚总数分为四个层级：tiny（<100K引脚）、small（100K–500K引脚）、medium（500K–5M引脚）和large（>5M引脚）。每个案例文件以纯文本格式存储，每行表示一个网络，格式为“NET <net_id> { <cell_id> <cell_id> ... }”。数据集适用于超图划分算法的基准测试，特别是在电子设计自动化（EDA）领域。Converted benchmark dataset for two-way hypergraph partitioning (min-cut under balance constraint). Derived from the integrated benchmark collection at Zenodo 15373339, re-formatted into a plain unweighted netlist format suitable for EDA partitioning toolchains. The dataset contains 489 cases with cell count ranging from 5,804 to 9,128,020, net count from 163 to 6,920,306, pin count from 6,323 to 57,156,537, and average degree from 1.45 to 1,825. Original format is hMETIS .hgr files, converted to unweighted netlist format. Cases are grouped into four tiers by total pin count: tiny (<100K pins), small (100K–500K pins), medium (500K–5M pins), and large (>5M pins). Each case file is in plain-text format with one net per line: NET <net_id> { <cell_id> <cell_id> ... }. It is intended for benchmarking hypergraph partitioning algorithms, particularly in electronic design automation (EDA).