Learning curves for an SVR trained using a feature vector of ﹩\left[{BP}\mbox{--}{RP},{RP}\mbox{--}H,{\rm{W}}1\mbox{--}{\rm{W}}2,J\mbox{--}H,J\mbox{--}K\right]﹩. A cross-validation set is separated out and the algorithm is trained on incrementally increasing training set size. The rms error (RMSE) score is then computed at each stage for both the training and validation set. Both curves tend toward each other smoothly to a sufficient accuracy of ∼0.17 dex. The good convergence and low final accuracy indicate the learning scenario has sufficiently low bias and variance.