20250902 meeting

Published 2025-08-29 Updated 2025-08-31 471 words 3 minutes views

Contents

Introduction

This experiment stems from modifying the LSTM part of the Regression Ensemble. By adjusting the parameters of the LSTM model, the goal is to make the model more accurate.

Code

The modified code is as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
    rnn_model = RNNModel(
        model="LSTM",
        input_chunk_length=90,
        hidden_dim=100,
        n_rnn_layers=5,
        dropout=0.1,
        training_length=120,
        batch_size=16,
        n_epochs=500,
        random_state=random_state,
        force_reset=True,
        save_checkpoints=False
    )

Simulation 1

Used for comparison with 20250821 meeting.

Method	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	8.290791581	8.156691128	9.101066219	141.46141777	141.23102811	142.85350026	2.963966534	2.833717649	3.750968858
RMSPE	2.879373470	2.855992144	3.016797345	11.89375541	11.88406614	11.95213371	1.721617418	1.683364978	1.936741815
MSPE%	0.027696563	0.027232927	0.030497993	0.47205142	0.47117129	0.47736946	0.009922369	0.009475392	0.012623134
RMSPE%	0.166422844	0.165024018	0.174636746	0.68705999	0.68641918	0.69091928	0.099611088	0.097341625	0.112352724
MAPE	1.297126883	1.277670082	1.414690631	10.64315946	10.63509135	10.69190938	0.923285580	0.903373232	1.043601881
MAPE%	0.004323483	0.004256935	0.004725586	0.03554397	0.03551039	0.03574685	0.003074664	0.003006797	0.003484735

Simulation 2

Used for comparison with 20250820 meeting.

Method	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	3.21773771	3.22087836	3.19876099	62.7561588	62.841457	62.2407622	0.83620087	0.83605521	0.83708095
RMSPE	1.79380537	1.79468057	1.78850804	7.9218785	7.927260	7.8892815	0.91444019	0.91436055	0.91492128
MSPE%	0.27116842	0.27160954	0.26850307	5.2483297	5.259654	5.1799036	0.07208197	0.07208775	0.07204705
RMSPE%	0.52073834	0.52116172	0.51817282	2.2909233	2.293394	2.2759402	0.26848086	0.26849162	0.26841582
MAPE	0.89236684	0.89236015	0.89240724	5.6202164	5.622268	5.6078195	0.70325285	0.70316383	0.70379076
MAPE%	0.07828959	0.07830305	0.07820826	0.4858846	0.486249	0.4836828	0.06198579	0.06198521	0.06198928

Epilogue

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1140819%20meeting/To%20be%20continued.jpg — To be continued!

Environment

Local Operating System: Windows 11 24H2
- Programming Language: Python 3.12.9
Computing Platform: National Center for High-Performance Computing (NCHC) – Taiwan AI Cloud
- Operating System: Ubuntu
- Miniconda
- GPU: NVIDIA Tesla V100 32GB GPU
- CUDA 12.8 driver
- Programming Language: Python 3.10.16 for Linux

Further Learning

I tested this project on the GitHub repository.

References

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) [Dataset]. Goddard Earth Sciences Data and Information Services Center (GES DISC). Retrieved from 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. Retrieved from https://openreview.net/forum?id=hHiIbk7ApW
SSSD (2022). GitHub. Retrieved from https://github.com/AI4HealthUOL/SSSD
SSSD_CP (2024). GitHub. Retrieved from https://github.com/egpivo/SSSD_CP
Unit8 SA (n.d.). Time Series Made Easy in Python. Darts. Retrieved from https://unit8co.github.io/darts/index.html
darts (2025). GitHub. Retrieved from https://github.com/unit8co/darts
Tzeng, S., & Huang, H. C. (2018). Resolution Adaptive Fixed Rank Kriging. Technometrics, 60(2), 198–208. Retrieved from https://doi.org/10.1080/00401706.2017.1345701
autoFRK (2024). GitHub. Retrieved from 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. Retrieved from https://arxiv.org/abs/2303.06053