From: Deep convolutional architectures for extrapolative forecasts in time-dependent flow problems
Hyperparameters | Values |
---|---|
Sequence length (\(n_t\)) | 5, 10, 20 |
LSTM layers | 1, 2, 3 |
hidden/latent dimension (m) | 25, 50, 125 |
Activation | tanh |
Loss function | MSE |
Learning rate | \(5 \times 10^{-4}\) |
Max model parameters (\(n_t = 20\)) | 126,000 |
Min model parameters (\(n_t = 20\)) | 5200 |