[Research method using discriminant subsets on minicomputers]

Given a set of elements (illnesses, medicines,...) characterised by criteria or attributes (symptoms, characteristics, colour,...), each able to assume a certain number of states (for colour, more than nine), the problem is to find a minimal discriminant chain, or in other words, the smallest sequence of criteria enabling one to distinguish the elements of the initial set, unless of course they are identical. This problem is very general, but when tackled together it often involves excessive calculations. Based on the research principles proposed by Begon and Tremolieres, the authors propose a method well adapted to criteria in several states, which can be used on a small computer, when searching for pseudo-minimal chains by sampling. An application of this method (in liaison with the 'Centre Anti-Poisons', Nancy Pr Larcan) has enabled us to carry out a search for discriminant chains for more than 2800 medical tablets by their external characteristics, among which 3 quantitative criteria were known with a certain margin of error, and 21 qualitative criteria, making a total of more than 60 different states.