DeepONet Project Update: Model Accessibility and Checkpoint Restoration Fix

[parsa-rahbari-82]

Description:

This pull request provides a series of important updates to the DeepONet project, primarily focusing on improving model accessibility and bug fixes. The following changes have been made:

1. Command-Line Model Parameters:

   • Added command-line arguments to give users more control over the model configuration without needing to modify the source code directly.
   • Users can now adjust critical parameters such as learning rate, number of sensors, and training epochs via the command line using Python's argparse library.

   Example usage:

   python deeponet_pde.py -p ode -m 200 --num-train 10000 --num-test 3000--lr 1e-3 --epochs 50000 --nn deeponet --activation relu --init Glorot --stacked False

2. Checkpoint Restoration Bug Fix:

   • Resolved a bug related to restoring checkpoints, which was caused by an incorrect file path.
   • The model can now properly restore from saved states for both training and inference.

3. Legendre Transform Flexibility:

   • Replaced the hardcoded time value (T=2) with a dynamic time variable.
   • This update ensures that the model can work with any time interval, providing flexibility and preventing extrapolation errors.

4. Compatibility with DeepONet’s Dynamic Systems:

   • Updated the structure of X_train and ensured proper formatting for X_test to be compatible with TensorFlow's compat.v1 backend.
   • Added necessary changes to avoid type mismatch errors, such as casting arrays in X_test to np.float32.

   Example code snippet:

X_test = tuple([arr.astype(dtype=np.float32) for arr in X_test])

Testing:

These changes have been thoroughly tested and confirmed to work reliably with TensorFlow's compat.v1 backend.

MIND Lab:

• Supervisor: Dr. Mostafa Abbaszadeh
  Team:
  • AmirAli Fakhari Zavareh
  • Melika Taheri Soufi
  • Parsa Rahbari

[parsa-rahbari-82]

Dear Dr. Lu
Thanks for the review! I’ve addressed the points you raised:

• Redundant python 2 removed.
• .gitignore file removed.
• Using original order of import.