Skip to content

Latest commit

 

History

History
26 lines (16 loc) · 1.78 KB

cuMultiFuzzyBaseline.md

File metadata and controls

26 lines (16 loc) · 1.78 KB

cuMultiFuzzyBaseline

These implementations accompany the paper “cuMFFP-Miner: GPU-Accelerated Multiple Fuzzy Frequent Pattern Mining via Shared Base-Item Traversal” by Mithun Thangaraj and Rage Uday Kiran, The University of Aizu, Japan.

Multiple fuzzy frequent pattern mining keeps all frequent fuzzy regions of an item, such as 1.L and 1.H, available for different patterns. Here, 1 is the base item and L or H is its fuzzy region. A pattern contains at most one region of each base item. A pattern is frequent when the sum of its minimum membership in each transaction reaches minSup.

cuMultiFuzzyBaseline is the flat GPU baseline. It processes region combinations as separate traversal jobs, which can repeat transaction matching for regions of the same base item.

Input is a fuzzy CSV/text file with rows such as 1.L\t2.H:0.8\t0.6. Labels contain an item and region separated by a dot; membership values range from 0 to 1. Support is the sum of minimum memberships across matching transactions.

minSup is an integer absolute fuzzy support or a float fraction in (0, 1]. sep selects the item delimiter. These miners require a local file. device selects the CUDA device, inputMode selects auto, gpu, or python parsing, and membershipScale controls membership precision (default 10). The default duplicate policy keeps the maximum membership per label.

Mining requires CuPy and an NVIDIA CUDA GPU.

from cupami import cuMultiFuzzyBaseline

miner = cuMultiFuzzyBaseline("fuzzy.csv", minSup=0.01)
miner.mine()
print(miner.getPatterns())
miner.save("patterns.txt")

getPatterns() maps tuples of region labels to fuzzy support values. save() writes patterns to a text file.

Use getLoadingTime(), getMiningTime(), and getTotalRuntime() for timings in seconds.