Table 4 Summary of hyperspaces for the data-driven models investigated in this work. For the GP, MLP, and XGB models, the optimization algorithm (see Sect. 3.5) searches these spaces over 100 iterations to identify the most performant hyperparameter set for each model
From: Machine learning applied to simulations of collisions between rotating, differentiated planets
Method | Hyperparameter | Range |
|---|---|---|
MLP | Number of layers | ∈{1,2,3} |
Neurons per layer | ∈{1,2,…,24} | |
GP | Kernel | Constant, Matérn 3/2, rational quadratic, radial-basis functions |
Noise (α) | ∈[0,10−2] | |
Kernel restart | ∈{0,1,…,5} | |
XGB | Number of estimators | ∈{1,10,…,1000} |
Maximum tree depth | ∈{3,4,…,12} | |
Column subsample ratio | ∈{0.5,…,1} | |
PCE | Polynomial order | ∈{2,3,…,15} |
q-norm | ∈{0.5,0.6,…,1.0} | |
Maximum interaction | ∈{2,3,…,5} | |
Feature importance | =0.01 |