chore: update dependencies

[Platohykj]

This pull request introduces several improvements to ensure compatibility with recent versions of Python, PyTorch, and scikit-learn, and enhances device management and data loading throughout the AttentionXML codebase. The most significant changes include adding allow_pickle=True to all relevant np.load calls, updating PyTorch and scikit-learn API usage for backward and forward compatibility, and improving device handling for both CPU and GPU environments.

Compatibility and Device Management Updates:

• Added allow_pickle=True to all np.load calls across the codebase to support loading objects saved with pickling, preventing errors with newer numpy versions. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
• Improved device management by dynamically selecting CUDA or CPU throughout the model and module classes, ensuring the code runs correctly regardless of hardware availability. [1] [2] [3] [4]

API and Backward Compatibility Fixes:

• Updated PyTorch API usage to use the new alpha and value keyword arguments in tensor operations, and ensured correct handling of .item() for scalar tensors. [1] [2] [3]
• Modified scikit-learn MultiLabelBinarizer usage to handle changes in module paths and initialization, improving backward compatibility. [1] [2]

Model and Data Pipeline Improvements:

• Updated embedding initialization and data handling in modules.py for better compatibility with PyTorch 2.x and improved padding/masking logic. [1] [2] [3] [4]
• Refined data loading and preprocessing functions to ensure robust handling of vocabularies and datasets. [1] [2] [3]

Documentation and Installation:

• Updated the README.md with a new installation section, including instructions for installing the latest version of PyTorch and using a requirements.txt file, replacing the old requirements list.

These changes collectively modernize the codebase, improve its robustness across environments, and make it easier to set up and run.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.