Click here for more information on the competition, which involved predicting the phase map of a three-component system using active learning. Each experiment returned realistic small-angle neutron scattering (SANS) data that varied with sample composition.
The first submission employed spectral clustering for the label phase, a Gaussian process classifier with a rational quadratic kernel for the extrapolate phase and entropy for the acquire phase.
The two alternative submissions made use of a KNN classifier for the extrapolate phase, respectively with n_neighbors=8 and 5.
