Table 3 Hartemink’s discretisation... | Emerald Publishing

Table 3

Hartemink’s discretisation algorithms

Input	A data set D of continuous variables, k₁ – that is, the initial (large) number of discretisation intervals – and k₂, the final discretisation intervals
Output	A discrete data set of factor variables, each discretised in k₂ levels in total
Initialisation
Discretise each variable independently using quantile discretisation and a large number k₁ of initial intervals
Specify the desired number of states of the variables in the end k₂
	Fork = k₁; k₂ + 1; k = k − 1 do
	Fori ← 1 to Ndo
	Compute pairwise mutual information coefficients:
	$M_{X_{i}^{k}} = \sum_{j \neq i} M I (X_{i}^{k}, X_{j}^{k})$
	Foreach pair l of adjascent intervals of X_ido
	Collapse each pairlof adjacent intervals of X_i in a single interval, and from the resulting variable $X_{i}^{k - 1} (l)$ compute:
	$M_{X_{i}^{k - 1} (l)} = \sum_{j \neq i} M I (X_{i}^{k - 1} (l), X_{j})$
	End
	Set $X_{i}^{k - 1} = \underset{X_{i}^{k - 1} (l)}{argmax} M_{X_{i}^{k - 1} (l)}$
	End
	End