Fig. 1From: JPlogP: an improved logP predictor trained using predicted dataRandom sampling versus targeted sampling. If there is a minimum learning requirement of 3 examples the random only is able to use 7 columns, while the targeted is able to learn on 18 columnsBack to article page