What is TensorPlay?
A learner-friendly deep learning framework built to be opened and changed.
Core goals
- Transparency: every abstraction can be peeled back.
- Simplicity: readable implementations come first.
- Entry point: familiar Python modules and training workflows lower the migration barrier.
- Hackability: experiment with hardware, operators, gradients, and graph optimization.
Four core libraries
P10 handles tensor computation, TPX handles autograd, Stax handles static graph optimization, and NN provides high-level network components. They follow one-way dependencies and can be used independently.