NEWS R Documentation

### Changes in FLSSS version 9.0.0 (20220415)

• Implemented multidimensional exact subset sum in arbitrary precision and magnitude, arbFLSSS(), and its distributed computing counterpart, decomposeArbflsss() and arbFLSSSobjRun(). A simple Bloom filter is added as an option to accelerate the algorithms.

### Changes in FLSSS version 8.6.0 (20200905)

• Implemented the distributed computing infrastructure, decomposeMflsss() and mFLSSSobjRun(), for multidimensional subset sum.

• Changed the timer from measuring total CPU time to measuring the elapsed wall time for multithreaded functions.

### Changes in FLSSS version 8.5.6 (20191027)

• Bug fix. A typo in the code related to integer type selection resulted in crash of the multidimensional subset sum miner given superset size > 127.

### Changes in FLSSS version 8.5.5 (20190709)

• Bug reported and fixed for FLSSS::FLSSS(). At a certain step, the algorithm uses a relative error threshold to decide if two floats are equal. Due to limited numeric precision, in an unusual instance the algorithm ended up comparing two subset sums 1e-14 and 1.2e-14. These two sums should both equal 0 theoretically (superset has negative elements). A wrong decision was thus made upon the significant relative error between 1e-14 and 1.2e-14. Bug fix consists of shifting superset v to nonnegative domain and scaling error bound ME to 1 before mining. The rest functions are not affected by this incident.

### Changes in FLSSS version 8.5 (20190106)

• Bug fix. The previous branch-and-bound knapsack solver did not initialize containers with sufficient sizes. In rare cases a pointer would be realized at the end of the container.

• Added a genetic algorithm with local heuristics for generalized assignment problem.

### Changes in FLSSS version 8.3 (20190106)

• Implemented different algorithms for the binary knapsack and the generalized assignment problems. Functions prefixed with 'aux' (auxiliary) are new.

• Future updates will dedicate to implementing existing foundational or state-of-the-art algorithms for combinatorial problems of interest. The implementation will focus on computing speed as usual.

### Changes in FLSSS version 7.7 (20181121)

• Integerization for the generalized assignment solver is canceled. A new compact representation for cost matrix is implemented and accelerates mining speed.

• Bug fix to the output format of GAP() given no solution exists.

• Package vignette is pushed to arXiv.