Giters

GestaltCogTeam / STEP

Code for our SIGKDD'22 paper Pre-training-Enhanced Spatial-Temporal Graph Neural Network For Multivariate Time Series Forecasting.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Error for running the code

Jimmy-7664 opened this issue · comments

commented

I've got several errors while running the code.
a).
For the forecasting part,

  1. I run the following command as written in the Readme.md "python step/run.py --cfg='step/step_METR-LA.py' --gpus='0'" then I got
    2022-09-29 15:15:25,915 - easytorch-launcher - INFO - Launching EasyTorch training.
    Traceback (most recent call last):
    File "step/run.py", line 20, in
    launch_training(args.cfg, args.gpus)
    File "/home/ght/tsformer/STEP/basicts/launcher.py", line 19, in launch_training
    easytorch.launch_training(cfg=cfg, gpus=gpus, node_rank=node_rank)
    File "/home/ght/anaconda3/envs/torch/lib/python3.8/site-packages/easytorch/launcher/launcher.py", line 58, in launch_training
    cfg = init_cfg(cfg, node_rank == 0)
    File "/home/ght/anaconda3/envs/torch/lib/python3.8/site-packages/easytorch/config/utils.py", line 210, in init_cfg
    cfg = import_config(cfg, verbose=save)
    File "/home/ght/anaconda3/envs/torch/lib/python3.8/site-packages/easytorch/config/utils.py", line 173, in import_config
    cfg = import(path, fromlist=[cfg_name]).CFG
    ModuleNotFoundError: No module named 'step.step_METR-LA'
  2. There seems only one dataset's scripts (METR-LA), no scripts for others like PEMS04 ets.

b).
For the pretraining part,
I run "python step/run.py --cfg='step/TSFormer_METR-LA.py' --gpus='0,1,2,3'",
then I got

2022-09-29 15:05:58,680 - easytorch-training - ERROR - Traceback (most recent call last):
File "/home/ght/anaconda3/envs/torch/lib/python3.8/site-packages/easytorch/launcher/launcher.py", line 30, in training_func
runner.train(cfg)
File "/home/ght/anaconda3/envs/torch/lib/python3.8/site-packages/easytorch/core/runner.py", line 361, in train
self.on_epoch_end(epoch)
File "/home/ght/tsformer/STEP/basicts/runners/base_runner.py", line 141, in on_epoch_end
if self.test_data_loader is not None and epoch % self.test_interval == 0:
AttributeError: 'TSFormerRunner' object has no attribute 'test_data_loader'

I check the scripts, it seems that test_data_loader is not defined.
How can I solve the above issues, looking forward to your reply.
: )

commented

Thanks for your question~

  1. This is a typo. The command line is case sensitive, so the correct command should be python step/run.py --cfg='step/STEP_METR-LA.py' --gpus='0'.
  2. STEP has just been refactored, and I'm testing other datasets (PEMS-BAY and PEMS04) to make sure there are no performance issues. However, my computing resources are limited, so it may take a few days. I will upload scripts for the PEMS04 and PEMS-BAY datasets very soon.
  3. I can not reproduce this error in the latest version of STEP. The test_data_loader is built in here. Could you provide more information?
commented

For the third question, I added the initialization of test_data_loader to avoid possible problems after you edit the config file (e.g., remove all arguments related to the test process).

commented

Thanks for your timely reply. Due to the limitation of computing resources, I will try again tomorrow.

commented

Testing on all these datasets has now been completed.
You can now find configurations for all datasets.
Moreover, our training logs are shown in training_logs/TSFormer_METR-LA.log, training_logs/TSFormer_PEMS04.log, and training_logs/TSFormer_PEMS04.log, and the our pre-trained TSFormers for each datasets are placed in tsformer_ckpt folder.