This is an implementation of ECLAT algorithm for frequent itemset mining in C language which uses bitset compression for the vertical transaction data. It can use one of four different bitset compression algorithms:
- Roaring bitmaps
- Bitmagic
- CONCISE
- EWAH
The source code is developed for the purpose of experimets of a paper presented at the International Congress on High-Performance Computing and Big Data Analysis (TopHPC 2019). For more information, please refer to the published paper:
Fadishei H., Doustian S., Saadati P. (2019) The Merits of Bitset Compression Techniques for Mining Association Rules from Big Data. In: Grandinetti L., Mirtaheri S., Shahbazian R. (eds) High-Performance Computing and Big Data Analysis. TopHPC 2019. Communications in Computer and Information Science, vol 891. Springer, Cham
Powerpoint presentation of the paper can be found here:
The selected bitset library can be specified at compile time using a BITSET environment variable. Its value can be one of ROARING, BM, EWAH, or CONCISE. For example:
make clean
BITSET=ROARING make
Run ./eclat -h to see the list of command line arguments.
./eclat -h
usage: eclat [options]
options:
-d <dataset> dataset file. csv of numbers. one transaction per line
-f <frac> fraction of transactions to process from start. default 1.0
-h print help
-H print header
-m <sup> minimum support. default 0.1
-p print frequent patterns
-s print stats
-v be verbose
The input file should be in the dat transaction format. Each line represents a transaction in which items are mentioned as 0-indexed integer values separated by space character.