Algorithm 2.2 Feature-wise NCM
Input: data examples T = {x1,…, xl+1} ⊂ X = Rk
Input: xj ∈ T (one of examples above)
Input: Background set U = {xl+2,…, xl+s+1} ⊂ X.
for i in 1, 2,…, k do
let ai be percentage of x1(i),…, xl+1(i) larger or equal than xj(i), or percentage of x1(i),…, xl+1(i) smaller or equal than xj(i), (the smallest one of two, taking values from 1/(l + 1) to 1/2);
let $a_i^{\prime }$ be percentage of xl+2(i),…, xl+s+1(i); xj(i) larger or equal than xj(i), or percentage of xl+2(i),…, xl+s+1(i) smaller or equal than xj(i), (the smallest one of two);
end for
Output: a NCM (ranged is from −k log(l + 1) to + k log(l + 1)) $${\alpha }_j=-\sum _{i=1}^k\text{log}(a_j/a_i^{\prime })$$