Marcin Raniszewski

Reference Set Reduction Algorithms Based on Double Sorting

The reference set reduction algorithm for 1-NN rule, resulting in the consistent reduced reference sets, is presented. The algorithm uses well known Hart’s algorithm. Samples from the reference set are initially sorted by two keys. The first one is the largest number of nearest neighbours from the same class, while the second one is a mutual distance measure introduced by Gowda and Krishna. The simple artificial example is used to show how the proposed algorithm operates. Two sets known as phoneme and satimage data have been used to verify the algorithm effectiveness, which is compared with the one offered by Gowda-Krishna algorithm.

FM-index for Dummies

The FM-index is a celebrated compressed data structure for full-text pattern searching. After the first wave of interest in its theoretical developments, we can observe a surge of interest in practical FM-index variants in the last few years. These enhancements are often related to a bit-vector representation, augmented with an efficient rank-handling data structure. In this work, we propose a new, cache-friendly, implementation of the rank primitive and advocate for a very simple architecture of the FM-index, which trades compression ratio for speed. Experimental results show that our variants are 2--3 times faster than the fastest known ones, for the price of using typically 1.5--5 times more space.

Sequential reduction algorithm for nearest neighbor rule

An effective training set reduction is one of the main problems in constructing fast 1-NN classifiers. A reduced set should be significantly smaller and ought to result in a similar fraction of correct classifications as a complete training set. In this paper a sequential reduction algorithm for nearest neighbor rule is described. The proposed method is based on heuristic idea of sequential adding and eliminating samples. The performance of the described algorithm is evaluated and compared with three other well-known reduction algorithms based on heuristic ideas, on four real datasets extracted from images.

The Sequential Reduction Algorithm for Nearest Neighbor Rule Based on Double Sorting

An effective and strong reduction of large training sets is very important for the Nearest Neighbour Rule usefulness. In this paper, the reduction algorithm based on double sorting of a reference set is presented. The samples are sorted with the use of the representative measure and Mutual Neighbourhood Value proposed by Gowda and Krishna. Then, the reduced set is built by sequential adding and removing samples according to double sort order. The results of proposed algorithm are compared with results of well-known reduction procedures on nine real and one artificial datasets.

Rank and select: Another lesson learned

Rank and select queries on bitmaps are essential building bricks of many compressed data structures, including text indexes, membership and range supporting spatial data structures, compressed graphs, and more. Theoretically considered yet in 1980s, these primitives have also been a subject of vivid research concerning their practical incarnations in the last decade. We present a few novel rank/select variants, focusing mostly on speed, obtaining competitive space-time results in the compressed setting. Our findings can be summarized as follows:

A Bloom filter based semi‐index on q‐grams

(i)

Sampled suffix array with minimizers.

no single rank/select solution works best on any kind of data (ours are optimized for concatenated bit arrays obtained from wavelet trees for real text datasets),

Prototype Extraction of a Single-Class Area for the Condensed 1-NN Rule

(ii)

Sampling the Suffix Array with Minimizers

it pays to efficiently handle blocks consisting of all 0 or all 1 bits,

The edited nearest neighbor rule based on the reduced reference set and the consistency criterion

(iii)