Catching Tensor Shape Errors without Running Your Code

Discover how to catch tensor shape mismatches in machine learning code without execution in this 27-minute PyCon US talk. Learn about representing symbolic tensor shapes using explicit type annotations called shape types and leveraging type checkers to identify errors. Explore the benefits of shape types for faster code comprehension through IDE integration. Gain insights into gradual adoption strategies for existing ML projects, support for broadcasting in NumPy and PyTorch, and understand the limitations of this innovative approach. Enhance your ML development workflow by reducing iteration times and simplifying debugging processes for both novice and experienced developers.

Talks - Pradeep Kumar Srinivasan: Catching Tensor Shape Errors without Running Your Code