Fig. 5
From: Tree-based learning for high-fidelity prediction of chaos
Prediction of Kuramoto–Sivashinsky equation with \(L=22\) and \(D = 64\) grid points. The x-axis shows Lyapunov time, where \(\lambda _{max}=0.043\). Using \(\xi =1\) and a quantile threshold of 0.5, TreeDOX selected \(k=26\) according to Eq. 1. Using \({\bf FI}\) from ETR #1, \(p=585\) of \(kD=1,664\) possible features are greater than \(FI_0\). (a,b) The test and forecasted dynamics, respectively. (c) The difference between the test and forecasted dynamics. Here, there were 46,837 and 1,162 training and testing samples used.
