11-07, 15:20–16:00 (US/Eastern), Central Park West
scikit-learn is a machine-learning library for Python that uses NumPy and SciPy for numerical operations. Scikit-learn has its own compiled code for performance-critical computation written in C, C++, and Cython. The library primarily focuses on Cython for compiled code because it is easy to use and approachable. In this talk, we dive into many techniques scikit-learn employs to utilize Cython fully. We will cover features like using the C++ standard library within Cython, fused types, code generation with the Tempita engine, and OpenMP for parallelization.
Scikit-learn is a machine-learning library largely backed by NumPy and SciPy. Although NumPy and SciPy have many compiled operations, scikit-learn requires domain-specific operations for performance-critical code. scikit-learn's compiled code is mainly in Cython because of its ease of use and approachability. At scikit-learn, we push Cython to its limits by using many of its features to support machine learning use cases. In this talk, we cover many of these techniques, such as:
- OpenMP to parallelize computation
- Fused types to write a function once and support multiple data types
- Using features from the C++ standard library such as vector, map, stack, or algorithm from Cython
- Memory-views to work with NumPy arrays with Python's Buffer protocol
- Tempita for code generation
- C structs and Python classes from Cython
By the end of this talk, you will have a deeper understanding of Cython's capabilities that you can apply to your use cases. This intermediate talk is designed for machine learning engineers, data scientists, and software engineers.
Previous knowledge expected
Thomas J. Fan is a senior machine learning engineer at Union.ai and a maintainer of scikit-learn, an open-source machine learning library for Python. He led the development of scikit-learn's set_output API, which allows transformers to return pandas DataFrames. Previously, Thomas worked at Columbia University to improve interoperability between scikit-learn and AutoML systems. He also maintains skorch, a neural network library that wraps PyTorch.