Skip to main content

Advertisement

Table 2 Costs chosen by the cross-validation for linear SVM using LIBLINEAR for the different training set sizes

From: Large-scale ligand-based predictive modelling using support vector machines

Solubility logD
Training set size Found cost Training set size Found cost
100 100,000 100 10,000
2 0.01
0.1 0.005
1000 0.05 1000 10,000,000
0.05 1000
0.05 0.1
5000 0.05 5000 0.5
0.05 0.75
0.05 0.1
10,000 0.05 10,000 100
0.05 2
0.05 0.25
20,000 0.1 20,000 0.5
0.1 0.5
0.1 1
32,096 0.1 80,000 0.5
0.1 0.5
0.1 0.75
  160,000 0.75
  0.5
  0.5
  320,000 0.75
  0.5
  0.5
  1,188,343 0.5
  0.5
  1
  1. Note the highly variable results among the three replicates for the small dataset sizes and low variation among the replicates for the larger training set sizes