Fig. 11

Ratio between the performance of ML and the best value output by TrivialRand for different instance groups and different training sets. TrivialRand is executed \(\min \{x(I),1000\}\) times for each instance I, x(I) being the number of runs that could be completed in the same time as one run of ML on I. The results are then averaged within each group. Each blue line represents the results obtained training the model with a different set of 10 randomly generated LGT networks with at most 100 leaves each. The green line corresponds to the training set used in [19]; the orange line represents one of the best-performing sets; the red line corresponds to the training set we used for the experiments on LGT and ZODS data in this paper, consisting of 1000 randomly generated LGT networks