Fig. 4
From: Large-scale ligand-based predictive modelling using support vector machines
Learning curves and model creation time for the solubility dataset. The plot shows learning curves (in black) overlaid with curves for model creation times (in red) for varying dataset sizes and the two SVM implementations. Plotted are measured values with lines drawn between the medians. πSVM was run on 64 cores on the cluster and LIBLINEAR was run using one such core. LIBLINEAR was much faster but even though the two black curves seem to converge there is a difference between them that in many cases probably is relevant, especially so at these small dataset sizes