Fig. 10

Log-probability of placing a consecutive sequence of t nodes correctly, starting from an initial partial tree with k nodes, for each model and dataset. We also plot the difference in log-probabilities between TGM and the random baseline model, where blue regions show our model outperforming the baseline and red regions highlight the opposite case. Note that \(t+k\) is equal to the tree size, so that tree size is conserved on the diagonals of these plots