doc: gpopt

Sequential optimization using GP algorithms


 [queries, Yt] = gpopt(f, Xt, Yt, Xs, T)
 [queries, Yt] = gpopt(..., 'Name',Value)


  • f function which returns observations
  • Xt matrix (nt, d) of initial data
  • Yt vector (nt, 1) of initial observations
  • Xs matrix (ns, d) of search data
  • T integer for the number of iterations

Name-Value Pair Arguments

  • algo string code for the query algorithm possible values are 'gpucb' (default), 'chaining', 'ei' for purely sequential optimization and 'gpucbpe', 'greedyucb' for batch sequential optimization
  • noise scalar for the noise standard deviation (default: 1e-2)
  • u scalar for the negative logarithm of the upper bound probability (default: 3)
  • B integer for the size of the batch in batch sequential optimization (default: 1)
  • kernel kernel function for the inferance (default: @kernel_se_normiso)
  • basis basis function for the inferance (default: @basis_none)
  • Kss matrix (ns,ns) of the kernel between points of Xs, used by the 'chaining' algorithm
  • plot boolean to visualize the optimization (default: false)
  • verbose boolean to monitor the optimization (default: true)


  • queries vector (1,T) of queries indices
  • Yt vector (T,1) of observations

See also

gpucb | chaining_ucb