Developer’s Guide

• Contributing
  • Ways to contribute
  • Submitting a bug report or a feature request
    • How to make a good bug report
  • Contributing code
    • Video resources
    • How to contribute
    • Pull request checklist
      • Continuous Integration (CI)
      • Stalled pull requests
      • Stalled and Unclaimed Issues
    • Issues for New Contributors
  • Documentation
    • Building the documentation
    • Guidelines for writing documentation
    • Generated documentation on CircleCI
  • Testing and improving test coverage
    • Writing matplotlib related tests
    • Workflow to improve test coverage
  • Monitoring performance
  • Issue Tracker Tags
  • Maintaining backwards compatibility
    • Deprecation
    • Change the default value of a parameter
  • Code Review Guidelines
    • Communication Guidelines
  • Reading the existing code base
• Developing scikit-learn estimators
  • APIs of scikit-learn objects
    • Different objects
    • Estimators
      • Instantiation
      • Fitting
      • Estimated Attributes
      • Optional Arguments
      • Pairwise Attributes
      • Universal attributes
  • Rolling your own estimator
    • get_params and set_params
    • Parameters and init
    • Cloning
    • Pipeline compatibility
    • Estimator types
    • Specific models
    • Estimator Tags
  • Coding guidelines
    • Input validation
    • Random Numbers
• Developers’ Tips and Tricks
  • Productivity and sanity-preserving tips
    • Folding and unfolding outdated diffs on pull requests
    • Checking out pull requests as remote-tracking branches
    • Display code coverage in pull requests
    • Useful pytest aliases and flags
    • Standard replies for reviewing
  • Debugging memory errors in Cython with valgrind
  • Building and testing for the ARM64 platform on a x86_64 machine
• Utilities for Developers
  • Validation Tools
  • Efficient Linear Algebra & Array Operations
  • Efficient Random Sampling
  • Efficient Routines for Sparse Matrices
  • Graph Routines
  • Testing Functions
  • Multiclass and multilabel utility function
  • Helper Functions
  • Hash Functions
  • Warnings and Exceptions
• How to optimize for speed
  • Python, Cython or C/C++?
  • Profiling Python code
  • Memory usage profiling
  • Performance tips for the Cython developer
    • Using OpenMP
  • Profiling compiled extensions
    • Using yep and gperftools
    • Using gprof
    • Using valgrind / callgrind / kcachegrind
      • kcachegrind
  • Multi-core parallelism using joblib.Parallel
  • A simple algorithmic trick: warm restarts
• Installing the development version of scikit-learn
  • Installing nightly builds
  • Building from source
    • Dependencies
      • Runtime dependencies
      • Build dependencies
      • Test dependencies
    • Building a specific version from a tag
    • Editable mode
  • Platform-specific instructions
    • Windows
    • macOS
      • macOS compilers from conda-forge
      • macOS compilers from Homebrew
    • Linux
      • Linux compilers from the system
      • Linux compilers from conda-forge
    • FreeBSD
  • Alternative compilers
    • Building with Intel C Compiler (ICC) using oneAPI on Linux
  • Parallel builds
• Bug triaging and issue curation
  • Working on issues to improve them
  • Working on PRs to help review
  • Triaging operations for members of the core and triage teams
  • A typical workflow for triaging issues
• Maintainer / core-developer information
  • Releasing
    • Before a release
    • Preparing a release PR
      • Major version release
      • Minor version release
    • Making a release
    • Release checklist
  • Merging Pull Requests
  • The scikit-learn.org web site
  • Travis Cron jobs
  • Experimental features
• Developing with the Plotting API
  • Plotting API Overview
  • Plotting with Multiple Axes