srush/Tensor-Puzzles
Tensor Puzzles
Tensor-Puzzles 是一套教育练习和数值计算教程，旨在掌握 PyTorch 中的张量运算和广播规则。它作为一个实现训练器，用户通过重新实现深度学习数学原语来练习将数学公式转换为代码。
该项目利用一套基于约束的练习，限制可用的库调用以强制使用特定的张量原语。这些挑战被结构化为顺序谜题，要求用户使用模块化实现模式解决任务，其中复杂函数被分解为更简单的依赖运算。
通过集成的 PyTorch 执行环境确保正确性，该环境使用参考实现验证和数值容差检查。系统验证用户定义的输出是否与参考结果匹配，并遵守标准的多元数组广播规则。
Features
- Tensor Broadcasting - Provides a mechanism for validating that multi-dimensional arrays align according to standard tensor broadcasting rules.
- Runtime Evaluation Environments - Provides an execution environment that runs user code within a live PyTorch session for validation.
- PyTorch Tensor Operations - Integrates with PyTorch to execute tensor operations and validate runtime shapes and types.
- PyTorch Code Exercises - Offers practical exercises and constrained challenges for learning multi-dimensional array manipulation in PyTorch.
- Machine Learning Mathematics - Teaches the fundamental mathematical operations and logic underlying neural network layers.
- Automated Exercise Suites - Provides a series of coding tasks bundled with automated verification for tensor operation implementation.
- API Call Restrictions - Restricts available library calls to force the use of specific tensor primitives during exercises.
- Reference Implementation Validation - Evaluates user submissions against a hidden gold-standard implementation to ensure mathematical equivalence.
- Implementation Trainers - Trains users to transition from mathematical formulas to efficient PyTorch code through verified exercises.
- Progressive Learning Paths - Structures tensor manipulation tasks as a series of progressive puzzles where solutions build on previous concepts.
- Programming Course Materials - Offers structured sets of interactive lessons for reimplementing deep learning mathematical primitives.
- Precision-Tolerant Comparisons - Uses epsilon-based precision-tolerant comparisons to handle floating point rounding errors in tensor results.
- Interactive Tutorials - Serves as a guided set of challenges for understanding tensor broadcasting and linear algebra fundamentals.
- Runtime Shape Verification - Provides runtime verification to ensure tensor dimensions adhere to broadcasting alignment rules.
- Numerical Tolerance Checks - Implements a test runner that verifies tensor output correctness using numerical tolerance epsilon checks.
