liuxu77 / LargeST

LargeST: A Benchmark Dataset for Large-Scale Traffic Forecasting (NeurIPS 2023 DB Track)
MIT License
166 stars 14 forks source link

There is something wrong about the results. Loss, RMSE, MAPE all is zero. #7

Closed Hevislee closed 3 months ago

Hevislee commented 3 months ago

Log directory: ./experiments/lstm/GLA/ 2024-07-24 14:05:42,361 - Namespace(device='cuda:0', dataset='GLA', years='2019', model_name='lstm', seed=2024, bs=32, seq_len=12, horizon=12, input_dim=3, output_dim=1, mode='train', max_epochs=100, patience=30, init_dim=32, hid_dim=64, end_dim=512, layer=2, lrate=0.001, wdecay=0.0001, dropout=0.1, clip_grad_value=5) 2024-07-24 14:05:55,062 - Data shape: (105120, 3834, 3) 2024-07-24 14:06:07,772 - Sample num: 63058, Batch num: 1970 2024-07-24 14:06:20,460 - Sample num: 21019, Batch num: 656 2024-07-24 14:06:33,156 - Sample num: 21020, Batch num: 656 2024-07-24 14:06:34,185 - The number of parameters: 97932 2024-07-24 14:06:34,185 - Start training! Check mask value tensor(0) 2024-07-24 14:13:28,434 - Epoch: 001, Train Loss: 0.0000, Train RMSE: 0.0000, Train MAPE: 0.0000, Valid Loss: 0.0000, Valid RMSE: 0.0000, Valid MAPE: 0.0000, Train Time: 310.5966s/epoch, Valid Time: 103.6521s, LR: 1.0000e-03 2024-07-24 14:13:28,438 - Val loss decrease from inf to 0.0000 2024-07-24 14:20:24,661 - Epoch: 002, Train Loss: 0.0000, Train RMSE: 0.0000, Train MAPE: 0.0000, Valid Loss: 0.0000, Valid RMSE: 0.0000, Valid MAPE: 0.0000, Train Time: 312.5177s/epoch, Valid Time: 103.7053s, LR: 1.0000e-03 2024-07-24 14:27:27,512 - Epoch: 003, Train Loss: 0.0000, Train RMSE: 0.0000, Train MAPE: 0.0000, Valid Loss: 0.0000, Valid RMSE: 0.0000, Valid MAPE: 0.0000, Train Time: 317.2444s/epoch, Valid Time: 105.6064s, LR: 1.0000e-03 2024-07-24 14:34:31,354 - Epoch: 004, Train Loss: 0.0000, Train RMSE: 0.0000, Train MAPE: 0.0000, Valid Loss: 0.0000, Valid RMSE: 0.0000, Valid MAPE: 0.0000, Train Time: 320.2488s/epoch, Valid Time: 103.5923s, LR: 1.0000e-03 2024-07-24 14:41:35,001 - Epoch: 005, Train Loss: 0.0000, Train RMSE: 0.0000, Train MAPE: 0.0000, Valid Loss: 0.0000, Valid RMSE: 0.0000, Valid MAPE: 0.0000, Train Time: 317.7777s/epoch, Valid Time: 105.8682s, LR: 1.0000e-03

liuxu77 commented 3 months ago

Hi, could you please provide more information, e.g., the environment, what you have modified.

Hevislee commented 3 months ago

Hi, could you please provide more information, e.g., the environment, what you have modified.

I’ve solved the problem, because the torch version. Thanks for your excellent work.

Hevislee commented 3 months ago

Have solved.