CMLA

doc: gpucb

Compute the ucb as in the GP-UCB algorithm

Syntax

 ucb = chaining_ucb(s2, u, n)

Arguments

  • s2 matrix (n, 1) of posterior variance
  • u scalar for negative log probability
  • n number of test points

Outputs

  • ucb vector (n, 1) such that $P[\forall x,~f(x)-\mu(x)>ucb(x)] < \exp(-u)$

See also

chaining_ucb