DyND: Enabling complex analytics across the language barrier
summary
A C++ library that actually implements a C++ array container with a full type system and lightweight Python bindings is presented, supporting variable-length strings, missing-value semantics via type-specific sentinels, ragged dimensions, structs and JSON-like data. Callables are first-class and publishable across languages, plugins load as shared libraries, and CUDA memory spaces are supported. Demos show elementwise, reduction and multi-dispatch functions working and benchmarks claim parity with NumPy for basic arithmetic while remaining experimental but usable.
AI-written summary · about
chapters (7)
- 0:00Introduction to Dined
- 1:16Purpose and design philosophy
- 4:20Solving array computing gaps
- 8:30Key traits and features
- 13:51Data types and metadata
- 20:34Functional programming in Dined
- 27:27Conclusion and Q&A
related talks
25m"Boosting Python Numeric Computations with Dynamic C++ Integration"
39mAccelerated Data Science Libraries
30mCuPy: A NumPy-compatible Library for GPU
27mData Apis: Standardization of N-dimensional arrays and dataframes
58mPython Data Ecosystem: Thoughts on Building for the Future
33mPortable, type-fancy multidimensional arrays
20mNumS: Scalable Array Programming for the Cloud
43mAnalyzing 3D objects with power of Deep Learning and Cython