Figure 4. Determining optimal number of NN layers and ensemble size. (left) Model prediction accuracy (on the NSTX-U validation dataset) across potential architectures. Each model consisted of a varying number of hidden layers, each with the same number of nodes per layer. An ensemble of 3 models was used for each case. (right) Prediction accuracy (on the NSTX-U validation dataset) as a function of number of models in the ensemble (for model with 4 layers of 140 nodes each).