About STEP-DCRNN

[Jimmy-7664]

I noticed that step-dcrnn experiments appear in the ablation experiments section, how did you incorporate step into dcrnn? A code example would be helpful.

[zezhishao]

The specific method is to fuse the representation generated by TSFormer with the Encoder of DCRNN, and the others are the same as Graph WaveNet.

[zezhishao]

Now I don't have the code too, I will reproduce it later.

[zezhishao]

See branch dev/dcrnn.
This is the result of the first 50 epochs on the METR-LA dataset:

2023-07-29 05:33:08,260 - easytorch-training - INFO - Result <train>: [train_time: 312.44 (s), lr: 1.00e-04, train_MAE: 2.7313, train_RMSE: 5.4791, train_MAPE: 0.0720]
2023-07-29 05:33:08,261 - easytorch-training - INFO - Start validation.
2023-07-29 05:34:07,963 - easytorch-training - INFO - Result <val>: [val_time: 59.70 (s), val_MAE: 2.7455, val_RMSE: 5.2874, val_MAPE: 0.0748]
2023-07-29 05:34:09,407 - easytorch-training - INFO - Checkpoint checkpoints/STEP_100/506878ab44e96728aec144de7098d6cb/STEP_best_val_MAE.pt saved
2023-07-29 05:36:09,617 - easytorch-training - INFO - Evaluate best model on test data for horizon 1, Test MAE: 2.1745, Test RMSE: 3.7863, Test MAPE: 0.0521
2023-07-29 05:36:09,621 - easytorch-training - INFO - Evaluate best model on test data for horizon 2, Test MAE: 2.4548, Test RMSE: 4.5666, Test MAPE: 0.0612
2023-07-29 05:36:09,625 - easytorch-training - INFO - Evaluate best model on test data for horizon 3, Test MAE: 2.6419, Test RMSE: 5.0937, Test MAPE: 0.0681
2023-07-29 05:36:09,628 - easytorch-training - INFO - Evaluate best model on test data for horizon 4, Test MAE: 2.7905, Test RMSE: 5.5172, Test MAPE: 0.0742
2023-07-29 05:36:09,632 - easytorch-training - INFO - Evaluate best model on test data for horizon 5, Test MAE: 2.9136, Test RMSE: 5.8632, Test MAPE: 0.0794
2023-07-29 05:36:09,637 - easytorch-training - INFO - Evaluate best model on test data for horizon 6, Test MAE: 3.0203, Test RMSE: 6.1615, Test MAPE: 0.0839
2023-07-29 05:36:09,641 - easytorch-training - INFO - Evaluate best model on test data for horizon 7, Test MAE: 3.1131, Test RMSE: 6.4187, Test MAPE: 0.0879
2023-07-29 05:36:09,644 - easytorch-training - INFO - Evaluate best model on test data for horizon 8, Test MAE: 3.1937, Test RMSE: 6.6351, Test MAPE: 0.0914
2023-07-29 05:36:09,648 - easytorch-training - INFO - Evaluate best model on test data for horizon 9, Test MAE: 3.2655, Test RMSE: 6.8282, Test MAPE: 0.0946
2023-07-29 05:36:09,652 - easytorch-training - INFO - Evaluate best model on test data for horizon 10, Test MAE: 3.3307, Test RMSE: 7.0007, Test MAPE: 0.0974
2023-07-29 05:36:09,655 - easytorch-training - INFO - Evaluate best model on test data for horizon 11, Test MAE: 3.3905, Test RMSE: 7.1538, Test MAPE: 0.1000
2023-07-29 05:36:09,658 - easytorch-training - INFO - Evaluate best model on test data for horizon 12, Test MAE: 3.4496, Test RMSE: 7.3009, Test MAPE: 0.1025
2023-07-29 05:36:09,762 - easytorch-training - INFO - Result <test>: [test_time: 120.35 (s), test_MAE: 2.9782, test_RMSE: 6.1186, test_MAPE: 0.0827]
2023-07-29 05:36:10,938 - easytorch-training - INFO - Checkpoint checkpoints/STEP_100/506878ab44e96728aec144de7098d6cb/STEP_034.pt saved

[Jimmy-7664]

Thank you for your prompt reply, my questions have been answered. Wish you all the best :)