run_tests.yaml

name: Run tests
on:
  pull_request:
  push: { branches: main }

jobs:
  run-test-suite:
    name: Run test suite
    runs-on: ubuntu-latest
    container: python:latest
      # TODO: use a gpu-compatible image, setup runners with a compatible gpu and activate
      # gpu passthrough options

    steps:
    - name: Checkout code
      uses: actions/checkout@v3

      # Use a pinned commit from the `feature/engine-api` branch at
      # https://github.com/scikit-learn/scikit-learn.git to enable smooth
      # synchronization with the development of this branch.
      # Development tracker: https://github.com/scikit-learn/scikit-learn/pull/25535/
      # TODO: Remove this step when the plugin API is officially released
    - name: Install pytest, sklearn branch "feature/engine-api", and sklearn-numba-dpex
      # Use official scikit-learn build guide at
      # https://scikit-learn.org/stable/developers/advanced_installation.html#install-bleeding-edge

      run: |
        apt-get update --quiet &&
        # Install prerequisites
        apt-get install -y build-essential python3-dev &&
        pip install cython numpy scipy joblib threadpoolctl &&
        # Build and install
        pip install torch --index-url https://download.pytorch.org/whl/cpu &&
        pip install pytest git+https://github.com/fcharras/scikit-learn.git@80f58bf10d2f8b8cb43f6253bbe13413985a1413#egg=scikit-learn -e .


    - name: Run sklearn_numba_dpex tests
      run: pytest -v sklearn_pytorch_engine/

      # TODO: run those tests in a separate pipeline
      # NB: `sklearn_pytorch_engine` set the estimators to output arrays of type
      # `torch.tensor` and store fitted attributes with this same type.
      # This behavior is not compatible with sklearn unit tests, that expect numpy
      # arrays or at least arrays that would closely mimic the NumPy Python API. To
      # keep compatibility with sklearn unit tests the engine must be set to a
      # different behavior where its methods are wrapped in data conversion steps so
      # that fitted attributes and outputs are numpy arrays. Currently this behavior is
      # activated when the environment variable SKLEARN_PYTORCH_ENGINE_TESTING_MODE is set
      # to 1.
    - name: Run sklearn test suites with sklearn_numba_dpex engines
      run: SKLEARN_RUN_FLOAT32_TESTS=1 SKLEARN_PYTORCH_ENGINE_TESTING_MODE=1 pytest -v --sklearn-engine-provider sklearn_pytorch_engine --pyargs sklearn.cluster.tests.test_k_means