For the purpose of our constructions, we will assume the existence of a reasonably large set so that each family
is locally finite and is a -locally finite base for a topology on F(X,L).
The metrics of F(X,L) can now be defined as constructions in [5] related to proving the metrization theorem:
where
and
It can be shown that is a (pseudo-)metric space for all . Since , it is intuitively clear that the 'gap' between U' and the complement of U closes up as r approaches s. We will use , to denote the intuitive counterpart of , where the 'gap' between U' and the complement of U is assumed to vanish or to be neglible.
Thus, we arrive at our proposal for a metric on F(X,L):
where is the cardinality for a set. For computing purposes we will assign to the following:
where , . This is a fair approximation, provided we consider and as 2-dimensional discs with radii . The activation is the non-overlapping area of the two circles normalized to the [0,1] interval. If and are identical, there is no non-overlapping area and the distance is zero. If the distance between and is more than there will be no overlapping area and reaches its maximum value.
The metric is based on an activation of which implies that if is a metric then is too. This property is guaranteed by the activation function.
We now have that is a pseudo-metric on F(X,L), and it is obvious that is not a metric, since this would imply that iff and the whole construction with would be of no use. The granularity of the pseudo-metric is provided by the size of the neighbourhoods within F(Y,M), as desired.