Algorithm 2.

t = buildExtraTree(X, Y)

if |X| < nmin OR isConstant(Y) then

Return buildLeaf(mean(Y))

end if

// Generate Candidate Thresholds at Random

{λ1, …, λn} ← GenerateSplits(min(X(j)),max(X(j))), ∀j ∈ {1, …, s}

// Select Threshold with Best Score (Eq. (4))

${\lambda }_j\leftarrow \underset{j}{\arg \max }{\psi }_{{\lambda }_j}\left(X\right)$

// Split Data, Recurse, and Build node

{Xl, Yl, Xr, Yr} ← λj(X, Y)

tl ← buildExtraTree(Xl, Yl)

tr ← buildExtraTree(Xr, Yr)

Return buildNode(λj, j, tl, tr)