Fig. 6From: Machine learning models for hydrogen bond donor and acceptor strengths using large and diverse training data generated by first-principles interaction free energiesViolin plots: tenfold internal cross validation results and test set performances (red dots) for various atomic reactivity descriptor types with their respective best sets of descriptor parameters trained on 4424 QC-HBA data points. RMSE, root mean square error. The descriptor abbreviations are as follows: CS charge shell; CRDF = charge radial distribution function, CACF = spatial charge autocorrelation function, MS mass shell; GACF topological charge autocorrelation function, combinatorial combination of CACF, CS, and shorted-shell. The mean RMSEs of the tenfold CV results are indicated above the descriptor abbreviations. The red dots mark the RMSE on the experimental test set. The partial charge type used was CM5 for all atoms in all casesBack to article page