Resource Efficiency for Time Series Forecasting on ECL
0.064Inference TimeFSMLP
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FSMLPSamples per inference=2562024.12 | 0.064 | — | — | — | — | — | |
| FreTSSamples per inference=2562024.12 | 0.069 | — | — | — | — | — | |
| TSMixerSamples per inference=2562024.12 | 0.081 | — | — | — | — | — | |
| iTransformerSamples per inference=2562024.12 | 0.121 | — | — | — | — | — | |
| DLinearInput length=96, Prediction length=96, Batch size=16, Number of layers=1, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 0.2 | 0.02 | 0.01 | 1.37 | 0.04 | 0.02 | |
| PatchTSTSamples per inference=2562024.12 | 0.207 | — | — | — | — | — | |
| FreTSInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 0.48 | 0.42 | 0.13 | 3.78 | 0.3 | 0.23 | |
| AutoformerSamples per inference=2562024.12 | 1.271 | — | — | — | — | — | |
| OLinearInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 2.11 | 4.79 | 1.65 | 7.75 | 0.45 | 0.17 | |
| vLinearInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 2.21 | 2.82 | 0.91 | 10.42 | 0.39 | 0.1 | |
| TimesNetSamples per inference=2562024.12 | 2.504 | — | — | — | — | — | |
| iTrans.Input length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 3.46 | 4.83 | 1.87 | 10.66 | 0.7 | 0.23 | |
| Fredf.m.Input length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 7.08 | 12.12 | 5.55 | 26.59 | 1.24 | 0.26 | |
| LeddamInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 14.78 | 8.56 | 5.32 | 65.19 | 1.44 | 0.49 | |
| PatchTSTInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 20.52 | 3.76 | 12.48 | 65.83 | 3.02 | 0.93 | |
| TimeMix.Input length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 22.79 | 0.12 | 1.74 | 57.09 | 4.24 | 0.82 | |
| CARDInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 28.21 | 1.39 | 5.07 | 78.25 | 5.08 | 0.63 | |
| TimeMix.++Input length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 34.1 | 1.19 | 64.02 | 106.06 | 1.02 | 0.36 | |
| TimesNetInput length=96, Prediction length=96, Batch size=16, Number of layers=2, FLOPs Unit=G, T.T. Unit=ms/iter, I.M. Unit=GB2026.01 | 157.13 | 300.58 | 293.98 | 509.17 | 5.82 | 1.41 |