Feat: computational graph

[ITHwang]

Implement basic tensor operations with autograd support

Summary

This PR implements a subset of PyTorch's tensor operations from scratch using NumPy as the computational backend. All operations support automatic differentiation through a define-by-run approach.

Features

• Core tensor creation functionality with torch.tensor()
• Basic arithmetic operations (+, -, *, /, negation)
• Mathematical functions (square(), exp(), pow())
• Full autograd support with backward propagation
• Type safety ensuring only floating-point tensors can require gradients
• Context manager (no_grad()) to disable gradient tracking for inference
• Memory optimization using weak references to prevent circular reference leaks

Testing

• Added comprehensive tests for all operations
• Numerical gradient checking for complex functions
• Verified memory optimization using before/after measurements

Implementation Notes

• Follows the Define-by-Run paradigm similar to PyTorch
• Operations are implemented as Function objects with forward and backward methods
• Each tensor stores references to its creator function for backpropagation

[Copilot]

Pull Request Overview

This PR introduces a draft implementation of a computational graph to support automatic differentiation using a new PriorityQueue for managing the backward propagation order. Key changes include:

• The addition of a generic PriorityQueue in torch/priority_queue.py for handling function ordering during backpropagation.
• Updates in torch/core.py to use this PriorityQueue and modifications to the backward propagation logic, including support for multi-branch operations.
• Enhancements in tests, documentation, and CI configuration to support and validate the new design.

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
torch/priority_queue.py	Adds a generic PriorityQueue implementation for function ordering.
torch/core.py	Updates the backward propagation logic using PriorityQueue; refactors Function implementations.
tests/torch/test_core.py	Adds a multi-branch backward test to ensure correct gradient propagation.
README.md	Expands documentation on core concepts and chain rule implementation.
.pre-commit-config.yaml	Updates configuration for pre-commit hooks.
.github/workflows/main.yml	Adjusts CI workflow to use pre-commit actions and install pytest.

Comments suppressed due to low confidence (2)

torch/core.py:256

• Duplicate backward method in the Exp class: the first backward definition that returns y should be removed to avoid confusion and unintended behavior, as the second definition properly handles input validation and gradient computation.

return y

torch/core.py:130

• The PriorityQueue class does not implement bool, so 'while funcs:' will always evaluate to True. Either implement a bool method for PriorityQueue or change the loop to 'while len(funcs):' to correctly terminate when the queue is empty.

while funcs:

[Copilot]

Copilot reviewed 13 out of 13 changed files in this pull request and generated 2 comments.