Table 1 The config of ViT-MAE.
From: Visual prediction method based on time series-driven LSTM model
Config | Value and Method |
|---|---|
Optimizer | AdamW |
Base learning rate | Le-3 |
Actual learning rate | 2.5e-4 |
Weight decay | 0.05 |
Optimizer momentum | \(\beta _{1}\), \(\beta _{2}\) =0.9, 0.95 |
Batch size | 64 |
Learning rate schedule | Cosine decay |
Loss function | MSE MAE |
Input length | 200 |
Patch size | \(16\times 16\) |
Encoder layers | 3 |
Dncoder layers | 1 |
Masking ratio | 85% |
