scientific article; zbMATH DE number 3856225

From MaRDI portal
Publication:3324855





We say that there is an inversion in a hierarchy if the height of a cluster merged later in the hierarchy is lower than that of clusters merged earlier. We give: 1) Necessary and sufficient conditions on the parameters of the G. C. Lance and W. T. Williams formula (1966), generalized by M. Jambu (1978) which insure the existence or non existence of inversion in a hierarchy; one of those new results generalizes the conditions given by P. Ducimetière (1971), G. Milligan (1979) and V. Batagelj (1981). 2) A necessary and sufficient condition on the same parameters which insures the neighbourhood reducibility property of M. Bruynooghe (1978) and thus permits the use of an accelerated algorithm for hierarchical clustering. By using those results and taking account of the a priori knowledge of the user we show that it is possible to obtain adaptive ultrametrics. This work arises towards learning hierarchical clustering from examples [see the author and \textit{J. V. Moreau}, Learning hierarchical clustering from examples. (1984)].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3324855)