Table 2 Statistical comparison between different RNN architectures.
From: pH trends and seasonal cycle in the coastal Balearic Sea reconstructed through machine learning
Slope | Intercept | Training error | Validation error | Training epochs | Training time | |
|---|---|---|---|---|---|---|
RNN | − 0.0021 ± 0.00077 | 8.07 ± 0.006 | 0.54 ± 0.08 | 0.72 ± 0.12 | 293 ± 95 | 15.52 ± 4.75 |
LSTM | − 0.0018 ± 0.00067 | 8.06 ± 0.005 | 0.49 ± 0.03 | 0.68 ± 0.05 | 245 ± 68 | 17.55 ± 4.21 |
BD-LSTM | − 0.0020 ± 0.00054 | 8.07 ± 0.004 | 0.46 ± 0.03 | 0.64 ± 0.04 | 167 ± 45 | 15.13 ± 3.00 |
BD-GRU | − 0.0020 ± 0.00066 | 8.07 ± 0.005 | 0.51 ± 0.07 | 0.74 ± 0.10 | 347 ± 95 | 27.68 ± 6.84 |
