1140821 meeting
前言
本次實驗同 1140819 meeting ,僅將訓練的資料集進行修改,嘗試簡化時空資料(#simulation01)。
設定檔
|
|
SSSD
+ autoFRK
autoFRK 推論耗時 2.17848 小時(CPU 平行運算,核心數:23)。
Metric | ALL Locs & All Time | Known Locs & All Time | Unknown Locs & All Time | ALL Locs & Future | Known Locs & Future | Unknown Locs & Future | ALL Locs & Past | Known Locs & Past | Unknown Locs & Past |
---|---|---|---|---|---|---|---|---|---|
MSPE | 3.015330438 | 2.892107713 | 3.759878621 | 3.149118440 | 3.024907082 | 3.899640239 | 3.009978918 | 2.886795739 | 3.754288156 |
RMSPE | 1.736470684 | 1.700619803 | 1.939040644 | 1.774575566 | 1.739226001 | 1.974750678 | 1.734929081 | 1.699057309 | 1.937598554 |
MSPE% | 0.010097198 | 0.009673286 | 0.012658600 | 0.010688266 | 0.010256705 | 0.013295887 | 0.010073555 | 0.009649949 | 0.012633108 |
RMSPE% | 0.100484813 | 0.098352863 | 0.112510443 | 0.103384072 | 0.101275392 | 0.115307794 | 0.100367101 | 0.098234153 | 0.112397101 |
MAPE | 0.955027571 | 0.939662570 | 1.047867454 | 1.036286040 | 1.024088108 | 1.109989559 | 0.951777232 | 0.936285549 | 1.045382569 |
MAPE% | 0.003180809 | 0.003127896 | 0.003500525 | 0.003494194 | 0.003451484 | 0.003752263 | 0.003168273 | 0.003114952 | 0.003490456 |
TSMixer
+ autoFRK
TSMixer 推論耗時 1:32:11.011375 (CPU 平行運算,核心數:24)。
autoFRK 推論耗時 3.103498 小時(CPU 平行運算,核心數:23)。
Metric | ALL Locs & All Time | Known Locs & All Time | Unknown Locs & All Time | ALL Locs & Future | Known Locs & Future | Unknown Locs & Future | ALL Locs & Past | Known Locs & Past | Unknown Locs & Past |
---|---|---|---|---|---|---|---|---|---|
MSPE | 2.978935565 | 2.849110373 | 3.763377811 | 3.353161356 | 3.23392849 | 4.07360163 | 2.963966534 | 2.833717649 | 3.750968858 |
RMSPE | 1.725959317 | 1.687930796 | 1.939942734 | 1.831163935 | 1.79831268 | 2.01831653 | 1.721617418 | 1.683364978 | 1.936741815 |
MSPE% | 0.009975697 | 0.009530008 | 0.012668681 | 0.011308897 | 0.01089540 | 0.01380736 | 0.009922369 | 0.009475392 | 0.012623134 |
RMSPE% | 0.099878410 | 0.097621759 | 0.112555238 | 0.106343298 | 0.10438104 | 0.11750471 | 0.099611088 | 0.097341625 | 0.112352724 |
MAPE | 0.930799848 | 0.911303942 | 1.048599886 | 1.118656559 | 1.10957169 | 1.17355002 | 0.923285580 | 0.903373232 | 1.043601881 |
MAPE% | 0.003100954 | 0.003034442 | 0.003502842 | 0.003758221 | 0.00372557 | 0.00395551 | 0.003074664 | 0.003006797 | 0.003484735 |
RegressionEnsemble
+ autoFRK
RegressionEnsemble 0:01:08.390425 (CPU 平行運算,核心數:12)。
autoFRK 推論耗時 2.511842 小時(CPU 平行運算,核心數:23)。
Metric | ALL Locs & All Time | Known Locs & All Time | Unknown Locs & All Time | ALL Locs & Future | Known Locs & Future | Unknown Locs & Future | ALL Locs & Past | Known Locs & Past | Unknown Locs & Past |
---|---|---|---|---|---|---|---|---|---|
MSPE | 2.982517642 | 2.852681404 | 3.767026630 | 3.446295352 | 3.326775292 | 4.168470940 | 2.963966534 | 2.833717649 | 3.750968858 |
RMSPE | 1.726996712 | 1.688988278 | 1.940882951 | 1.856420037 | 1.823944981 | 2.041683360 | 1.721617418 | 1.683364978 | 1.936741815 |
MSPE% | 0.009988315 | 0.009542616 | 0.012681360 | 0.011636978 | 0.011223224 | 0.014137010 | 0.009922369 | 0.009475392 | 0.012623134 |
RMSPE% | 0.099941559 | 0.097686316 | 0.112611546 | 0.107874828 | 0.105939718 | 0.118899140 | 0.099611088 | 0.097341625 | 0.112352724 |
MAPE | 0.931842171 | 0.912334128 | 1.049715542 | 1.145756950 | 1.136356529 | 1.202557080 | 0.923285580 | 0.903373232 | 1.043601881 |
MAPE% | 0.003104506 | 0.003037965 | 0.003506567 | 0.003850575 | 0.003817178 | 0.004052370 | 0.003074664 | 0.003006797 | 0.003484735 |
RegressionEnsemble
(LSTM) + autoFRK
RegressionEnsemble 推論耗時 0:22:01.921387 (CPU 平行運算,核心數:12)。
autoFRK 推論耗時 3.103747 小時(CPU 平行運算,核心數:23)。
Metric | ALL Locs & All Time | Known Locs & All Time | Unknown Locs & All Time | ALL Locs & Future | Known Locs & Future | Unknown Locs & Future | ALL Locs & Past | Known Locs & Past | Unknown Locs & Past |
---|---|---|---|---|---|---|---|---|---|
MSPE | 2.981051376 | 2.851161365 | 3.765885281 | 3.408172444 | 3.287254270 | 4.138795853 | 2.963966534 | 2.833717649 | 3.750968858 |
RMSPE | 1.726572146 | 1.688538233 | 1.940588901 | 1.846123626 | 1.813078672 | 2.034403070 | 1.721617418 | 1.683364978 | 1.936741815 |
MSPE% | 0.009984350 | 0.009538438 | 0.012678679 | 0.011533879 | 0.011114599 | 0.014067291 | 0.009922369 | 0.009475392 | 0.012623134 |
RMSPE% | 0.099921719 | 0.097664930 | 0.112599640 | 0.107395899 | 0.105425801 | 0.118605613 | 0.099611088 | 0.097341625 | 0.112352724 |
MAPE | 0.931999359 | 0.912484185 | 1.049915819 | 1.149843849 | 1.140258019 | 1.207764269 | 0.923285580 | 0.903373232 | 1.043601881 |
MAPE% | 0.003105247 | 0.003038683 | 0.003507442 | 0.003869825 | 0.003835849 | 0.004075121 | 0.003074664 | 0.003006797 | 0.003484735 |
結論
Method / Model | SSSD + autoFRK | TSMixer + autoFRK | RegressionEnsemble + autoFRK | RegressionEnsemble (LSTM) + autoFRK |
---|---|---|---|---|
MSPE ALL Locs (Future) | 3.149118440 | 3.353161356 | 3.446295352 | 3.408172444 |
MSPE Known Locs (Future) | 3.024907082 | 3.233928490 | 3.326775292 | 3.287254270 |
MSPE Unknown Locs (Future) | 3.899640239 | 4.073601630 | 4.168470940 | 4.138795853 |
RMSPE ALL Locs (Future) | 1.774575566 | 1.831163935 | 1.856420037 | 1.846123626 |
RMSPE Known Locs (Future) | 1.739226001 | 1.798312680 | 1.823944981 | 1.813078672 |
RMSPE Unknown Locs (Future) | 1.974750678 | 2.018316530 | 2.041683360 | 2.034403070 |
MSPE% ALL Locs (Future) | 0.010688266 | 0.011308897 | 0.011636978 | 0.011533879 |
MSPE% Known Locs (Future) | 0.010256705 | 0.010895400 | 0.011223224 | 0.011114599 |
MSPE% Unknown Locs (Future) | 0.013295887 | 0.013807360 | 0.014137010 | 0.014067291 |
RMSPE% ALL Locs (Future) | 0.103384072 | 0.106343298 | 0.107874828 | 0.107395899 |
RMSPE% Known Locs (Future) | 0.101275392 | 0.104381040 | 0.105939718 | 0.105425801 |
RMSPE% Unknown Locs (Future) | 0.115307794 | 0.117504710 | 0.118899140 | 0.118605613 |
MAPE ALL Locs (Future) | 1.036286040 | 1.118656559 | 1.145756950 | 1.149843849 |
MAPE Known Locs (Future) | 1.024088108 | 1.109571690 | 1.136356529 | 1.140258019 |
MAPE Unknown Locs (Future) | 1.109989559 | 1.173550020 | 1.202557080 | 1.207764269 |
MAPE% ALL Locs (Future) | 0.003494194 | 0.003758221 | 0.003850575 | 0.003869825 |
MAPE% Known Locs (Future) | 0.003451484 | 0.003725570 | 0.003817178 | 0.003835849 |
MAPE% Unknown Locs (Future) | 0.003752263 | 0.003955510 | 0.004052370 | 0.004075121 |
結語
運行環境
- 本機作業系統:Windows 11 24H2
- 程式語言:Python 3.12.9
- 計算平臺:財團法人國家實驗研究院國家高速網路與計算中心臺灣 AI 雲
- 作業系統:Ubuntu
- Miniconda
- GPU:NVIDIA Tesla V100 32GB GPU
- CUDA 12.8 driver
- 程式語言:Python 3.10.16 for Linux
延伸學習
- 我測試此項目的 Github 儲存庫 。
參考資料
Global Modeling and Assimilation Office (GMAO)。(2015)。MERRA-2 tavg1_2d_flx_Nx: 2d,1-Hourly,Time-Averaged,Single-Level,Assimilation,Surface Flux Diagnostics (Version 5.12.4) [資料集]。Goddard Earth Sciences Data and Information Services Center (GES DISC)。參考自 https://doi.org/10.5067/7MCPBJ41Y0K6
Juan Lopez Alcaraz 、 Nils Strodthoff(2022)。Diffusion-based time series imputation and forecasting with structured state space models。Transactions on Machine Learning Research。參考自 https://openreview.net/forum?id=hHiIbk7ApW
SSSD(2022)。GitHub。參考自 https://github.com/AI4HealthUOL/SSSD
SSSD_CP(2024)。GitHub。參考自 https://github.com/egpivo/SSSD_CP
Unit8 SA(無日期)。Time Series Made Easy in Python。Darts。參考自 https://unit8co.github.io/darts/index.html
darts(2025)。GitHub。參考自 https://github.com/unit8co/darts
Tzeng, S., & Huang, H. C. (2018). Resolution Adaptive Fixed Rank Kriging. Technometrics, 60(2), 198–208. 參考自 https://doi.org/10.1080/00401706.2017.1345701
autoFRK(2024)。GitHub。參考自 https://github.com/egpivo/autoFRK
Si-An Chen, Chun-Liang Li, Nate Yoder, Sercan O. Arik and Tomas Pfister. (2023). TSMixer: An all-MLP architecture for time series forecasting. arXiv. 參考自 https://arxiv.org/abs/2303.06053