Extended Data Fig. 5: Hyper-parameter selection for the random forests model.
From: Co-limitation towards lower latitudes shapes global forest diversity gradients
Using 10 bootstrapping iterations on random training sets consisting of 1000 random sample for each continent, we calculated the sensitivity of the root-mean-squared error (RMSE) to (left) the number of trees to grow and (right) the number of variables randomly sampled as candidates at each split in the random forests model. As RMSE converged at 50 trees and 10 variables, we selected them as optimal hyper-parameters for the random forests model. Black lines represent the mean of all the iterations (with red bands showing the standard error of the mean), and blues lines represent each iteration.
