MFITS : Bayesian active learning with support for continuous fidelity parameters and variable cost