Research activities

 

Bit string representation of analysed data

The core of the bit string representation of analysed data is shown at Fig. 1.

Fig. 1:  Cards of categories of attribute A1 of data matrix M
Rows of M Attributes of M Cards of categories of attribute A1
A1 A2 … AK A1[1] A1[2] A1[3] A1[4]
o1 1 6 … B 1 0 0 0
o2 2 4 … C 0 1 0 0
o3 1 7 … G 1 0 0 0
… … … … … … … … …
on - 1 4 9 … F 0 0 0 1
on 3 8 … H 0 0 1 0

Here we have data matrix M with attributes A1, A2, …, AK and with rows o1, …, on. We suppose that the {1, 2, 3, 4} are categories (i.e. possible values) of the attribute A1. Thus the attribute A1 is represented by cards A1[1], A1[2], A1[3], A1[4] of categories 1, 2, 3, 4 respectively.

The card A1[1] of the category 1 is the string of bits. Each row of data matrix M corresponds to one bit of the card A1[1]. There is “1” in the bit corresponding to the row oi if and only if there is the value 1 in the row oi. The same is true for other categories and attributes.

The cards of categories and bit string operations are used to compute cards of literals and cards of derived Boolean.

The card A[α] of literal A(α) is a string of bits representing literal A(α). It is e.g.

A[1,2,4] = A[1] A[2] A[4]

where A[1] A[2] A[4] is a bit-wise disjunction of cards of categories A[1], A[2] and A[4].

The card of Boolean attribute φ is a string of bits representing φ. E.g. the card of derived Boolean attribute A1(1,2) A2(3,4) is a conjunction A1[1,2] A2[3,4] of cards A1[1,2] and A2[3,4] of literals A1(1,2) and A2(3,4).

The bit-wise operations on cards are carried out by very fast computer instructions. The very fast computer instructions are used also in realisation of the function Count(ξ) that returns the number of “1” in the string of bits ξ. The function Count(ξ) is used in computation of necessary contingency tables.

Some further data structures and sophisticated algorithm based on cards of categories are used in analytical procedures of the system LISp-Miner. They approved to be fast enough for teaching and lot of research purposes, see e.g. nμco o 4ft, Florida,

This bit string approach was used already in 1971 (see [Ra 71]) by J. Rauch in the frame of development of GUHA method, see also [Ra 78]. It was adapted for the LISp-Miner and implemented and further developed by M. Šimωnek, see e.g. [RS 01B] and [Si 03].

The current research concerns application of this approach in new analytical procedures of the LISp-Miner system. The other direction of research concerns the application of the bit string approach in multi-relational data mining as suggested in [Ra 86], see also [Ra 02B].

Print page    PDF version

 
 

Send comments about this site to the webmaster