Table 2. Main features of chosen MLPs and CNNs.
| Model | Activationa | No. of fully connected layers (neurons) | No. of convolutional layers (filters) | No. SNPs/window (stride) | Dropout (weight regularization) |
|---|---|---|---|---|---|
| MLP1 | Elu | 1 (32) | NA | NA | 0.01 (0.0) |
| MLP2 | Elu | 2 (64) | NA | NA | 0.03 (0.0) |
| MLP3 | Softplus | 5 (32) | NA | NA | 0.01 (0.0) |
| MLP-hot | Elu | 4 (128) | NA | NA | 0.03 (0.01) |
| CNN1 | Linear | 1 (32) | 1 (16) | 3 (1) | 0.01 (0.0) |
| CNN2 | Elu | 3 (32) | 1 (32) | 2 (1) | 0.01 (0.0) |
| CNN3 | Softplus | 3 (64) | 1 (16) | 2 (1) | 0.01 (0.0) |
No., number; MLP, Multilayer Perceptron; Elu, exponential linear unit; CNN, Convolutional Neural Network.
a
Elu: f(x) = c (ex−1) x < 0, f(x) = x, x > 0; SoftPlus: f(x) = ln(1+ex); and Linear: f(x) = c x.