Table 3 Darts-based default hypertunning parameters for DNN models used in this study.
SN | Model name | Default hypertunning parameters in a darts-based environment | |
|---|---|---|---|
Input variable | Input action | ||
1. | N-BEATS | Number of stacks | 30 |
Generic architecture | True | ||
Number of layers | 4 | ||
Number of blocks | 1 | ||
Layer width | 256 | ||
Expansion coefficients dim | 5 | ||
Trend polynomial degree | 2 | ||
Dropout | 0.0 | ||
Activation function | Relu | ||
Output chunk shift | 0 | ||
2. | TCN | Output chunk shift | 0 |
Kernal size | 3 | ||
No of layers | None | ||
Dilation base | 2 | ||
Weight norm | False | ||
Dropout | 0.2 | ||
3. | RNN | Output chunk shift | 0 |
Model | “RNN” or “LSTM” or “GRU” | ||
Hidden dim | 25 | ||
RNN layer | 1 | ||
Dropout | 0.0 | ||
Hidden fc sizes | None | ||
Batch size | [32, 64] | ||
4. | Transformer | Output chunk shift | 0 |
Kernal size | 3 | ||
No of filter | 3 | ||
No of layer | None | ||
Dilation base | 2 | ||
Weight norm | False | ||
Dropout | 0.2 | ||
Random state | 0 | ||
