Official Journal of the Asia Oceania Geosciences Society (AOGS)
From: Streamlining hyperparameter optimization for radiation emulator training with automated Sherpa
Atmospheric condition | Radiation type | Geographic type | Number of hidden layers (H) | ||||
---|---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | |||
Clear sky | LW | Land | 436 | 379 | 231 | 219 | 166 |
Ocean | 966 | 119 | 249 | 184 | 168 | ||
SW | Land | 505 | 379 | 274 | 242 | 144 | |
Ocean | 130 | 79 | 138 | 156 | 148 | ||
Cloudy sky | LW | Land | 471 | 370 | 263 | 237 | 123 |
Ocean | 384 | 275 | 253 | 187 | 138 | ||
SW | Land | 443 | 217 | 190 | 234 | 206 | |
Ocean | 188 | 304 | 284 | 164 | 130 | ||
Sherpa-derived average neurons | 440.4 | 265.3 | 235.3 | 202.9 | 152.9 | ||
Speedup (fold) | 12.4 | 9.3 | 7.4 | 7.1 | 9.4 | ||
Reduction (%) | 18.1 | 13.5 | 10.7 | 10.3 | 13.6 |