katiana22 / tl-deeponet Goto Github PK

The test.m in your respository gives a exp operation to K_field, and in the dataset you give a np.log to conteract exp, does this have specific reason? I did not notice this and have generated some labels use the exp K_field...

I was trying to run TL1/source_model.py but I met an error at this line:

The error says:

Exception has occurred: ValueError
cannot reshape array of size 20000000 into shape (1541)
  File "~/TL-DeepONet/TL1/dataset.py", line 71, in load_data
    u_train_mean = np.mean(np.reshape(u_train, (-1, r)), 0)
  File "~/TL-DeepONet/TL1/dataset.py", line 17, in __init__
    self.X, self.u_mean, self.u_std = self.load_data()
  File "~/TL-DeepONet/TL1/source_model.py", line 74, in main
    data = DataSet(bs)
  File "~/TL-DeepONet/TL1/source_model.py", line 227, in <module>
    main()
ValueError: cannot reshape array of size 20000000 into shape (1541)

This is because u_train has shape (2000, 100, 100), but I guess it's supposed to be (2000, 1541), where 1541 is the number of output sensors.

The data I used is generated here:

Your repository I am working with utilizes TensorFlow 1.15 with the CPU version. However, I would like to configure the environment to run TensorFlow on GPU. I am encountering difficulties in setting up the environment to enable GPU acceleration for TensorFlow 1.15. The 3090 GPU was abandoned by me because it does not support versions below cuda11.0. And I attempted to configure this environment on a TITAN V GPU using cuda 10.0 and cudnn 7.4.2, based on the information provided on the TensorFlow official website. Even though tf.test.is_gpu_available() returned true, I still could not run your code correctly. Besides, I tried rewriting your code using PyTorch, but the test results on the TL7 source dataset differed significantly from the results in your paper. (2.12% in my experiment; 1.51% in your paper) If you could help me solve this problem, or provide the code for the GPU version, I would be very grateful.