1150310 meeting

收尾工作 v1.0

發佈於 2026-03-08 更新於 2026-03-09 約 3007 字預計閱讀 7 分鐘次閱讀

本次實驗複測 Weather2K 、 NASA GES DISC MERRA-2 、 2b_8 資料集。

使用的資料集大小如下：

資料集	變數數量	已知地點	`batch_size`	未知地點	訓練時間步	未來時間步
Weather2K	4	1,492	100	374	1,176	88
MERRA-2	1	950	100	50	4,608	144
2b_8	1	9,500	200	500	90	10

實驗設定

本次實驗使用設定如下：

config.yaml

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
wavenet:
  # WaveNet model parameters
  input_channels:  1  # Number of input channels
  output_channels: 1  # Number of output channels
  residual_layers: 32  # Number of residual layers
  residual_channels: 64  # Number of channels in residual blocks
  skip_channels: 64  # Number of channels in skip connections

  # Diffusion step embedding dimensions
  diffusion_step_embed_dim_input:  128  # Input dimension
  diffusion_step_embed_dim_hidden: 256  # Middle dimension
  diffusion_step_embed_dim_output: 256  # Output dimension

  # Structured State Spaces sequence model (S4) configurations
  s4_max_sequence_length: 90  # Maximum sequence length
  s4_state_dim: 128  # State dimension
  s4_dropout: 0.2  # Dropout rate
  s4_bidirectional: true  # Whether to use bidirectional layers
  s4_use_layer_norm: true  # Whether to use layer normalization

diffusion:
  # Diffusion model parameters
  T: 100  # Number of diffusion steps
  beta_0: 0.0001  # Initial beta value
  beta_T: 0.01  # Final beta value

AFRK:
# AutoFRK model parameters, for more details, please refer to https://pypi.org/project/autoFRK/
  method: "fast"  # autoFRK method to use (e.g., "fast")
  tps_method: "rectangular"  # autoFRK's TPS method to use (e.g., "rectangular")

其中，依資料集不同，修改的部分為 input_channels 、 output_channels 、 s4_max_sequence_length 。

在本次實驗中，所有資料集皆改為將訓練集直接填補未來幾期，經測試後發現可大幅提升填補的準確率。同時測試 s4_bidirectional 對填補的影響。針對 Weather2K 資料集，測試 s4_dropout 為 0.2 的情況；針對 MERRA-2 資料集，同時測試 s4_dropout 為 0.0 、 0.2 、 0.8 不同情況；針對 2b_8 資料集，測試 s4_dropout 為 0.0 的情況。

在 training.yaml 中，測試有無 enable_spatial_training（AFRK）的影響。

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
# Training configuration
batch_size: 200  # Batch size
output_directory: "/home/u6025091/SSSD_CP/results/2b_8-fast-rectangular"  # Output directory for checkpoints and logs
ckpt_iter: "max"  # Checkpoint mode (max or min)
iters_per_ckpt: 500  # Checkpoint frequency (number of epochs)
iters_per_logging: 200  # Log frequency (number of iterations)
n_iters: 38000  # Maximum number of iterations
learning_rate: 0.0005  # Learning rate

# Additional training settings
only_generate_missing: true  # Generate missing values only
use_model: 2  # Model to use for training
masking: "forecast"  # Masking strategy for missing values
missing_k: 10  # Number of missing values

# Data paths
data:
  train_path: "/home/u6025091/SSSD_CP/datasets/2b_8/data_train_known_real.npy"  # Path to training data

# autoFRK config
enable_spatial_training: true  # Enable spatial training step
location_path: "/home/u6025091/SSSD_CP/datasets/2b_8/stations_known_locations.npy"  # Path to known locations

在 inference.yaml 中，測試 enable_spatial_normalization（SP）的影響。

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
# Inference configuration
batch_size: 200  # Batch size for inference
output_directory: "/home/u6025091/SSSD_CP/results/2b_8-fast-rectangular/inference"  # Output directory for inference results
ckpt_path: "/home/u6025091/SSSD_CP/results/2b_8-fast-rectangular"  # Path to checkpoint for inference
trials: 1 # Replications

# Additional training settings
only_generate_missing: true  # Generate missing values only
use_model: 2  # Model to use for training
masking: "forecast"  # Masking strategy for missing values  # inference mode need to fix in the code, or using other masking strategy will be failed
missing_k: 10  # Number of missing values

# Data paths
data:
  test_path: "/home/u6025091/SSSD_CP/datasets/2b_8/data_train_known_real.npy"  # Path to test data

# autoFRK config
enable_spatial_inference: true  # Enable spatial prediction step
enable_spatial_normalization: false  # Enable spatial normalization before and after autoFRK
known_location_path: "/home/u6025091/SSSD_CP/datasets/2b_8/stations_known_locations.npy"  # Path to known locations
unknown_location_path: "/home/u6025091/SSSD_CP/datasets/2b_8/stations_unknown_locations.npy"  # Path to unknown locations

Weather2K

bidirectional-FRK-4000

尚未完成實驗。

bidirectional-NoFRK-4000

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	9.212197e+00	4.764740e+00	10.829105	8.104409e+01	68.246143	23.431347	3.837021e+00	1.443116e-02	9.886081
RMSPE	3.035160e+00	2.182828e+00	3.290761	9.002449e+00	8.261122	4.840594	1.958832e+00	1.201298e-01	3.144214
MSPE%	1.607277e+10	1.713991e+06	0.178218	6.444603e+10	1.965038	0.405369	1.245300e+10	1.842249e+06	0.161220
RMSPE%	1.267784e+05	1.309195e+03	0.422158	2.538622e+05	1.401798	0.636686	1.115930e+05	1.357295e+03	0.401522
MAPE	8.855900e-01	4.531534e-01	1.857254	5.544696e+00	5.434782	2.891528	5.369495e-01	8.037850e-02	1.779859
MAPE%	4.245987e+08	2.953705e+07	0.041180	1.013968e+09	0.161959	0.064585	3.804962e+08	3.174730e+07	0.039428

NoSP

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	9.622670e+00	5.138295e+00	13.595236	8.215610e+01	73.610913	26.586258	4.194998e+00	1.449347e-02	12.623119
RMSPE	3.102043e+00	2.266781e+00	3.687172	9.064000e+00	8.579680	5.156186	2.048169e+00	1.203888e-01	3.552903
MSPE%	1.585812e+10	2.447800e+06	0.245626	6.400976e+10	2.199272	0.484240	1.225494e+10	2.630969e+06	0.227770
RMSPE%	1.259290e+05	1.564545e+03	0.495606	2.530015e+05	1.482994	0.695874	1.107020e+05	1.622026e+03	0.477253
MAPE	9.472319e-01	4.569086e-01	2.333831	5.602686e+00	5.487968	3.326226	5.988646e-01	8.043473e-02	2.259571
MAPE%	4.190081e+08	3.820500e+07	0.055416	1.009678e+09	0.163989	0.077277	3.748083e+08	4.106387e+07	0.053780

Nobidirectional-FRK-4000

尚未完成實驗。

Nobidirectional-NoFRK-4000

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/2kNobidirectionalNoFRK4000log.png

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	8.557305e+00	4.357313e+00	10.717023	7.167319e+01	62.453320	21.991220	3.834348e+00	9.992905e-03	9.873375
RMSPE	2.925287e+00	2.087418e+00	3.273686	8.466002e+00	7.902741	4.689480	1.958149e+00	9.996452e-02	3.142193
MSPE%	1.532682e+10	4.910454e+06	0.175495	5.342669e+10	1.243336	0.367536	1.247581e+10	5.277903e+06	0.161125
RMSPE%	1.238015e+05	2.215954e+03	0.418921	2.311421e+05	1.115050	0.606248	1.116952e+05	2.297369e+03	0.401403
MAPE	8.404920e-01	4.067417e-01	1.847794	5.013504e+00	4.959049	2.774382	5.282258e-01	6.609286e-02	1.778458
MAPE%	4.162873e+08	5.306461e+07	0.041032	9.481803e+08	0.129175	0.062375	3.764858e+08	5.703544e+07	0.039435

NoSP

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	8.821512e+00	4.808902e+00	13.022649	7.165436e+01	68.935532	24.962443	4.119734e+00	1.031077e-02	12.129195
RMSPE	2.970103e+00	2.192921e+00	3.608691	8.464890e+00	8.302742	4.996243	2.029713e+00	1.015420e-01	3.482699
MSPE%	1.546526e+10	4.921911e+06	0.228755	5.539211e+10	1.372258	0.442306	1.247754e+10	5.290217e+06	0.212775
RMSPE%	1.243594e+05	2.218538e+03	0.478283	2.353553e+05	1.171434	0.665061	1.117029e+05	2.300047e+03	0.461275
MAPE	8.856481e-01	4.269254e-01	2.242633	5.031184e+00	5.235553	3.147337	5.754380e-01	6.709618e-02	2.174934
MAPE%	4.147628e+08	5.343217e+07	0.052522	9.579761e+08	0.136539	0.073606	3.741141e+08	5.743049e+07	0.050944

MERRA-2

bidirectional-FRK-4000

尚未完成實驗。

bidirectional-NoFRK-4000

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	0.631161	0.477201	10.280107	10.491658	8.513340	15.152535	0.066237	0.016798	10.000957
RMSPE	0.794456	0.690798	3.206261	3.239083	2.917763	3.892626	0.257365	0.129606	3.162429
MSPE%	0.002294	0.001739	0.035207	0.038344	0.031085	0.054628	0.000228	0.000057	0.034095
RMSPE%	0.047892	0.041698	0.187636	0.195817	0.176309	0.233727	0.015106	0.007581	0.184648
MAPE	0.251795	0.232853	2.906040	2.635178	2.543609	3.155217	0.115247	0.100466	2.891764
MAPE%	0.000897	0.000827	0.010018	0.009595	0.009263	0.011408	0.000399	0.000344	0.009938

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/MERRA2bidirectionalNoFRK4000SP1.png

NoSP

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	198236.030804	0.367509	3.879090e+06	158755.391217	6.478062	3.110934e+06	200497.942447	0.017425	3.923099e+06
RMSPE	445.237050	0.606225	1.969541e+03	398.441202	2.545204	1.763784e+03	447.769966	0.132003	1.980682e+03
MSPE%	688.413089	0.001336	1.351371e+04	576.070077	0.023611	1.133382e+04	694.849408	0.000060	1.363860e+04
RMSPE%	26.237627	0.036548	1.162485e+02	24.001460	0.153657	1.064604e+02	26.359996	0.007719	1.167844e+02
MAPE	80.254871	0.219046	1.583713e+03	74.552805	2.250021	1.426400e+03	80.581552	0.102688	1.592725e+03
MAPE%	0.278721	0.000776	5.517933e+00	0.270491	0.008189	5.196684e+00	0.279192	0.000351	5.536338e+00

bidirectional-NoFRK-dropout0.0-4000

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	0.679363	0.237607	10.461359	11.335908	4.042383	22.427100	0.068832	0.019625	9.775821
RMSPE	0.824235	0.487450	3.234402	3.366884	2.010568	4.735726	0.262358	0.140090	3.126631
MSPE%	0.002458	0.000855	0.035912	0.041218	0.014586	0.080938	0.000238	0.000068	0.033332
RMSPE%	0.049583	0.029233	0.189504	0.203021	0.120771	0.284496	0.015423	0.008239	0.182571
MAPE	0.259026	0.183983	2.926754	2.667272	1.587548	4.105072	0.121054	0.103570	2.859247
MAPE%	0.000923	0.000649	0.010100	0.009693	0.005751	0.014857	0.000421	0.000356	0.009828

NoSP

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	272856.376224	0.228059	5.315017e+06	242581.732982	3.865768	4.726952e+06	274590.860993	0.019648	5.348708e+06
RMSPE	522.356560	0.477555	2.305432e+03	492.525870	1.966156	2.174155e+03	524.014180	0.140173	2.312727e+03
MSPE%	948.153585	0.000819	1.855650e+04	880.933318	0.013936	1.728082e+04	952.004747	0.000068	1.862959e+04
RMSPE%	30.792103	0.028626	1.362223e+02	29.680521	0.118053	1.314565e+02	30.854574	0.008244	1.364902e+02
MAPE	88.936696	0.181412	1.752376e+03	88.691191	1.539571	1.702171e+03	88.950762	0.103600	1.755252e+03
MAPE%	0.309068	0.000639	6.114465e+00	0.321926	0.005576	6.209867e+00	0.308331	0.000356	6.108999e+00

bidirectional-NoFRK-dropout0.8-4000

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	1.124980	1.106429	10.146221	19.710693	20.270051	14.558338	0.060173	0.008513	9.893443
RMSPE	1.060651	1.051869	3.185313	4.439673	4.502227	3.815539	0.245302	0.092268	3.145384
MSPE%	0.004122	0.004039	0.034800	0.072441	0.074030	0.053435	0.000207	0.000029	0.033733
RMSPE%	0.064199	0.063555	0.186548	0.269149	0.272084	0.231161	0.014401	0.005416	0.183664
MAPE	0.281098	0.287881	2.894569	3.584130	4.110291	3.182689	0.091862	0.068889	2.878062
MAPE%	0.001010	0.001035	0.009987	0.013090	0.014978	0.011638	0.000318	0.000237	0.009892

NoSP

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	150574.444572	1.987353	2.942067e+06	133335.662610	36.514928	2.609129e+06	151562.083122	0.009211	2.961141e+06
RMSPE	388.039231	1.409735	1.715245e+03	365.151561	6.042758	1.615280e+03	389.309752	0.095974	1.720797e+03
MSPE%	522.973730	0.007246	1.025620e+04	484.079926	0.133161	9.536985e+03	525.202021	0.000032	1.029740e+04
RMSPE%	22.868619	0.085121	1.012729e+02	22.001816	0.364912	9.765749e+01	22.917287	0.005628	1.014761e+02
MAPE	68.112053	0.367612	1.342040e+03	67.630784	5.516283	1.270001e+03	68.139626	0.072636	1.346168e+03
MAPE%	0.236618	0.001324	4.678681e+00	0.245512	0.020084	4.633175e+00	0.236109	0.000249	4.681288e+00

Nobidirectional-FRK-4000

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/MERRA2NobidirectionalFRK4000log.png

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	0.934441	1.179018	9.580757	16.313171	21.751974	8.117971	0.053368	0.000359	9.664563
RMSPE	0.966665	1.085826	3.095280	4.038957	4.663901	2.849205	0.231016	0.018951	3.108788
MSPE%	0.003425	0.004299	0.032767	0.059999	0.079321	0.029589	0.000184	0.000001	0.032949
RMSPE%	0.058523	0.065569	0.181017	0.244947	0.281639	0.172015	0.013556	0.001117	0.181520
MAPE	0.219504	0.243107	2.817106	3.278275	4.273947	2.423857	0.044262	0.012174	2.839636
MAPE%	0.000794	0.000883	0.009709	0.011977	0.015561	0.008829	0.000153	0.000042	0.009760

NoSP

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	182,103.4469	1.0957	3,553,594.0	167,537.1967	20.2144	3,269,625.0	182,937.9717	0.0004	3,569,864.0
RMSPE	426.7358	1.0468	1,885.0980	409.3131	4.4960	1,808.2100	427.7125	0.0197	1,889.4080
MSPE%	632.8748	0.0040	12,403.5000	608.0813	0.0738	11,929.6600	634.2953	0.0000	12,430.6400
RMSPE%	25.1570	0.0632	111.3710	24.6593	0.2716	109.2230	25.1852	0.0012	111.4928
MAPE	78.4212	0.2340	1,546.8880	78.8324	4.0981	1,497.2510	78.3976	0.0126	1,549.7310
MAPE%	0.2725	0.0009	5.3968	0.2860	0.0149	5.4566	0.2717	0.0000	5.3934

Nobidirectional-NoFRK-4000

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/MERRA2NobidirectionalNoFRK4000log.png

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	0.651523	0.535322	10.326410	10.814419	9.535042	15.213563	0.069273	0.019713	10.046417
RMSPE	0.807169	0.731657	3.213473	3.288528	3.087886	3.900457	0.263198	0.140403	3.169608
MSPE%	0.002367	0.001950	0.035351	0.039531	0.034826	0.054848	0.000238	0.000067	0.034234
RMSPE%	0.048657	0.044163	0.188019	0.198824	0.186617	0.234197	0.015437	0.008180	0.185025
MAPE	0.260319	0.250812	2.906929	2.671431	2.760482	3.155861	0.122182	0.107029	2.892667
MAPE%	0.000926	0.000890	0.010019	0.009726	0.010057	0.011410	0.000422	0.000365	0.009940

NoSP

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	213455.299309	0.402967	4.175011e+06	198345.954454	7.097287	3.881056e+06	214320.938858	0.019438	4.191852e+06
RMSPE	462.012229	0.634797	2.043284e+03	445.360477	2.664073	1.970040e+03	462.948095	0.139420	2.047401e+03
MSPE%	741.021085	0.001465	1.454442e+04	719.588489	0.025888	1.414078e+04	742.248994	0.000066	1.456755e+04
RMSPE%	27.221702	0.038279	1.206003e+02	26.825147	0.160899	1.189150e+02	27.244247	0.008126	1.206961e+02
MAPE	81.883167	0.229040	1.614716e+03	81.173912	2.366525	1.555511e+03	81.923801	0.106580	1.618108e+03
MAPE%	0.284439	0.000810	5.628739e+00	0.294478	0.008618	5.665761e+00	0.283864	0.000363	5.626618e+00

Nobidirectional-NoFRK-10000

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	0.635422	0.484750	10.496749	10.497099	8.552034	17.694556	0.070430	0.022561	10.084375
RMSPE	0.797134	0.696240	3.239869	3.239923	2.924386	4.206490	0.265387	0.150205	3.175590
MSPE%	0.002305	0.001766	0.035962	0.038305	0.031243	0.063775	0.000243	0.000077	0.034369
RMSPE%	0.048014	0.042025	0.189637	0.195716	0.176757	0.252537	0.015583	0.008794	0.185388
MAPE	0.260637	0.249688	2.906929	2.616535	2.604560	3.449520	0.125664	0.114774	2.901471
MAPE%	0.000926	0.000886	0.010019	0.009519	0.009490	0.012471	0.000435	0.000393	0.009971

NoSP

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	181860.883779	0.452924	3.535570e+06	166326.115224	7.964999	3.232551e+06	182750.896561	0.022545	3.552931e+06
RMSPE	426.451502	0.672996	1.880311e+03	407.830989	2.822233	1.797930e+03	427.493739	0.150149	1.884922e+03
MSPE%	634.310666	0.001650	1.238971e+04	604.564722	0.029095	1.181983e+04	636.014861	0.000077	1.242236e+04
RMSPE%	25.185525	0.040616	1.113091e+02	24.587898	0.170572	1.087190e+02	25.219335	0.008791	1.114556e+02
MAPE	76.138943	0.243951	1.499120e+03	75.044246	2.497090	1.432609e+03	76.201660	0.114865	1.502931e+03
MAPE%	0.265154	0.000865	5.242707e+00	0.272554	0.009097	5.229755e+00	0.264730	0.000393	5.243449e+00

2b_8

bidirectional-FRK-s4_dropout0.0-4000

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/2b_8bidirectionalFRK4000log.png

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	5.8231e+66	0.1026	1.1646e+68	1.0083e+66	1.0260	2.0166e+67	6.3581e+66	0.000005	1.2716e+68
RMSPE	2.4131e+33	0.3203	1.0792e+34	1.0041e+33	1.0129	4.4906e+33	2.5215e+33	0.002124	1.1277e+34
MSPE%	1.4416e+73	0.1900	2.8833e+74	6.9926e+66	1.8993	1.3985e+68	1.6018e+73	0.000051	3.2036e+74
RMSPE%	3.7969e+36	0.4359	1.6980e+37	2.6443e+33	1.3782	1.1826e+34	4.0023e+36	0.007167	1.7899e+37
MAPE	3.6612e+32	0.0819	7.3224e+33	1.5208e+32	0.8097	3.0416e+33	3.8990e+32	0.000991	7.7980e+33
MAPE%	3.7969e+39	0.3450	7.5937e+40	1.0044e+33	3.3521	2.0088e+34	4.2187e+39	0.010899	8.4375e+40

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/2b_8bidirectionalFRK4000SP1.png

NoSP

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	5.823092e+66	0.102603	1.164618e+68	1.008290e+66	1.025985	2.016580e+67	6.358069e+66	0.000005	1.271614e+68
RMSPE	2.413108e+33	0.320316	1.079175e+34	1.004136e+33	1.012909	4.490635e+33	2.521521e+33	0.002124	1.127659e+34
MSPE%	1.441628e+73	0.189981	2.883256e+74	6.992550e+66	1.899346	1.398510e+68	1.601809e+73	0.000051	3.203617e+74
RMSPE%	3.796877e+36	0.435868	1.698015e+37	2.644343e+33	1.378168	1.182586e+34	4.002260e+36	0.007167	1.789865e+37
MAPE	3.661201e+32	0.081864	7.322402e+33	1.520802e+32	0.809727	3.041605e+33	3.899023e+32	0.000991	7.798046e+33
MAPE%	3.796872e+39	0.345023	7.5937e+40	1.004424e+33	3.352141	2.008849e+34	4.218747e+39	0.010899	8.437494e+40

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/2b_8bidirectionalFRK4000NoSP1.png

bidirectional-NoFRK-s4_dropout0.0-4000

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/2b_8bidirectionalNoFRK4000log.png

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	5.147689e+66	0.100314	1.029538e+68	2.716749e+65	0.994832	5.433498e+66	5.689469e+66	0.000923	1.137894e+68
RMSPE	2.268852e+33	0.316724	1.014661e+34	5.212244e+32	0.997413	2.330987e+33	2.385261e+33	0.030387	1.066721e+34
MSPE%	6.448289e+72	0.178323	1.289658e+74	1.978870e+66	1.743544	3.957740e+67	7.164765e+72	0.004410	1.432953e+74
RMSPE%	2.539348e+36	0.422283	1.135631e+37	1.406723e+33	1.320433	6.291057e+33	2.676708e+36	0.066404	1.197060e+37
MAPE	3.230288e+32	0.103238	6.460575e+33	7.896310e+31	0.797468	1.579262e+33	3.501472e+32	0.026101	7.002943e+33
MAPE%	2.539343e+39	0.486568	5.078687e+40	5.159410e+32	3.351772	1.031882e+34	2.821493e+39	0.168212	5.642985e+40

NoSP

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	5.888744e+66	0.100352	1.177749e+68	5.335852e+65	0.995214	1.067170e+67	6.483761e+66	0.000923	1.296752e+68
RMSPE	2.426673e+33	0.316785	1.085241e+34	7.304692e+32	0.997604	3.266758e+33	2.546323e+33	0.030388	1.138750e+34
MSPE%	5.439624e+73	0.178050	1.087925e+75	3.127449e+66	1.740829	6.254897e+67	6.044027e+73	0.004408	1.208805e+75
RMSPE%	7.375381e+36	0.421960	3.298371e+37	1.768459e+33	1.319405	7.908791e+33	7.774334e+36	0.066394	3.476788e+37
MAPE	3.693160e+32	0.103256	7.386319e+33	1.056481e+32	0.797643	2.112962e+33	3.986124e+32	0.026101	7.972248e+33
MAPE%	7.375381e+39	0.486633	1.475076e+41	6.438245e+32	3.352859	1.287649e+34	8.194868e+39	0.168163	1.638974e+41

Nobidirectional-FRK-s4_dropout0.0-4000

尚未完成實驗。

Nobidirectional-NoFRK-s4_dropout0.0-4000

尚未完成實驗。

實驗快照

https://raw.githubusercontent.com/Josh-test-lab/website-assets-repository/refs/heads/main/posts/1150310%20meeting/Experimental%20snapshot.png — 實驗快照。

參考資料

Zhu X, Xiong Y, Wu M, et al. Weather2K: A Multivariate Spatio-Temporal Benchmark Dataset for Meteorological Forecasting Based on Real-Time Observation Data from Ground Weather Stations[C]//International Conference on Artificial Intelligence and Statistics. PMLR, 2023: 2704-2722.
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