Add daily unit tests (#502)

.github/actions/daily_unittests/action.yml

@@ -0,0 +1,29 @@
+name: Run image
+description: Run private container from GitHub Packages Registry after Docker login
+inputs:
+  python_version:
+    description: 'Python version'
+    required: true
+  pandas_version:
+    description: 'Pandas version'
+    required: true
+  numpy_version:
+    description: 'Numpy version'
+    required: true
+  scikit_version:
+    description: 'Scikit-learn version'
+    required: true
+  scipy_version:
+    description: 'Scipy version'
+    required: true
+  tabmat_version:
+    description: 'Tabmat version'
+    required: true
+runs:
+  using: 'docker'
+  image: condaforge/mambaforge:latest
+  args:
+   - /bin/bash
+   - "-l"
+   - "-c"
+   - "PYTHON_VERSION=${{ inputs.python_version }} PANDAS_VERSION=${{ inputs.pandas_version }} NUMPY_VERSION=${{ inputs.numpy_version }} SCIKIT_VERSION=${{ inputs.scikit_version }} SCIPY_VERSION=${{ inputs.scipy_version }} TABMAT_VERSION=${{ inputs.tabmat_version }} ./.github/workflows/test_daily.sh"

.github/workflows/daily.yml

@@ -0,0 +1,72 @@
+name: Daily runs
+on:
+  schedule:
+    - cron: '0 5 * * *'
+  push:
+    paths:
+    - '.github/workflows/daily.yml'
+
+jobs:
+  linux-daily-unittests:
+    name: "Linux - daily unit tests - Python ${{ matrix.PYTHON_VERSION}} - ${{ matrix.NOTE }}"
+    runs-on: ubuntu-latest
+    env:
+      CI: True
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - PYTHON_VERSION: '3.9'
+            PANDAS_VERSION: 'nightly'
+            NUMPY_VERSION: 'nightly'
+            # See https://github.com/Quantco/glum/issues/506
+            # SCIKIT_VERSION: 'nightly'
+            SCIKIT_VERSION: ''
+            SCIPY_VERSION: 'nightly'
+            TABMAT_VERSION: 'nightly'
+            NOTE: 'Nightly Builds' # run once with nightlies
+          - PYTHON_VERSION: '3.9'
+            PANDAS_VERSION: ''
+            NUMPY_VERSION: ''
+            SCIKIT_VERSION: ''
+            SCIPY_VERSION: ''
+            TABMAT_VERSION: ''
+            NOTE: 'Default Builds' # run once with normal dependencies
+    steps:
+      - name: Pull image
+        run: docker pull condaforge/mambaforge:latest
+      - name: Checkout branch
+        uses: actions/checkout@v2.3.5
+        with:
+          ref: ${{ github.head_ref }}
+      - name: Run CI inside of container
+        uses: ./.github/actions/daily_unittests
+        with:
+          python_version: ${{ matrix.PYTHON_VERSION }}
+          pandas_version: ${{ matrix.PANDAS_VERSION }}
+          numpy_version: ${{ matrix.NUMPY_VERSION }}
+          scikit_version: ${{ matrix.SCIKIT_VERSION }}
+          scipy_version: ${{ matrix.SCIPY_VERSION }}
+          tabmat_version: ${{ matrix.TABMAT_VERSION }}
+      - name: Issue on failure
+        uses: actions/github-script@v5
+        if: ${{ failure() }}
+        with:
+          script: |
+            github.rest.issues.listForRepo({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              state: "open",
+              labels: "[bot] Daily run"
+            }).then((issues) => {
+              if (issues.data.length === 0){
+                github.rest.issues.create({
+                  owner: context.repo.owner,
+                  repo: context.repo.repo,
+                  title: "Daily run failure: Unit tests",
+                  body: "The daily unit tests failed. See https://github.com/Quantco/glum/actions/runs/${{ github.run_id }} for details.",
+                  assignees: ["MarcAntoineSchmidtQC", "lbittarello"],
+                  labels: ["[bot] Daily run"]
+                })
+              }
+            });

.github/workflows/test_daily.sh

@@ -0,0 +1,74 @@
+#!/bin/bash
+
+set -exo pipefail
+
+SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
+source ${SCRIPT_DIR}/base.sh "$PYTHON_VERSION"
+
+mamba install -y yq
+
+cat environment.yml > /tmp/environment.yml
+
+# pin version of some libraries, if specified
+LIBRARIES=("python" "pandas" "numpy" "scikit" "scipy" "tabmat")
+for library in "${LIBRARIES[@]}"; do
+    varname="${library^^}_VERSION"
+    version=${!varname}
+    if [[ -n "$version" && "$version" != "nightly" ]]; then
+        if [[ "${library}" == "scikit" ]]; then
+            library="scikit-learn"
+        fi
+        yq -Y --in-place ". + {dependencies: [.dependencies[], \"${library}=${version}\"]}" /tmp/environment.yml
+    fi
+done
+
+cat /tmp/environment.yml
+
+mamba env create -f /tmp/environment.yml
+mamba env update -n $(yq -r .name environment.yml) --file environment-benchmark.yml
+conda activate $(yq -r .name environment.yml)
+
+# This needs to be done before any 'uninstall --force'
+if [[ "$TABMAT_VERSION" == "nightly" ]]; then
+    echo "Install compilation dependencies"
+    mamba install -y c-compiler cxx-compiler cython jemalloc-local libgomp mako xsimd
+fi
+
+PRE_WHEELS="https://pypi.anaconda.org/scipy-wheels-nightly/simple"
+if [[ "$NUMPY_VERSION" == "nightly" ]]; then
+    echo "Installing Numpy nightly"
+    conda uninstall -y --force numpy
+    pip install --pre --no-deps --upgrade --timeout=60 -i $PRE_WHEELS numpy
+fi
+if [[ "$PANDAS_VERSION" == "nightly" ]]; then
+    echo "Installing Pandas nightly"
+    conda uninstall -y --force pandas
+    pip install --pre --no-deps --upgrade --timeout=60 -i $PRE_WHEELS pandas
+fi
+if [[ "$SCIKIT_VERSION" == "nightly" ]]; then
+    echo "Install scikit-learn nightly"
+    conda uninstall -y --force scikit-learn
+    pip install --pre --no-deps --upgrade --timeout=60 -i $PRE_WHEELS scikit-learn
+fi
+if [[ "$SCIPY_VERSION" == "nightly" ]]; then
+    echo "Installing Scipy nightly"
+    conda uninstall -y --force scipy
+    pip install --pre --no-deps --upgrade --timeout=60 -i $PRE_WHEELS scipy
+fi
+if [[ "$TABMAT_VERSION" == "nightly" ]]; then
+    echo "Install tabmat nightly"
+    # TODO: switch to a special channel once we have a Quetz instance up and running
+    conda uninstall -y --force tabmat
+    pip install git+https://github.com/Quantco/tabmat
+fi
+
+pip install --no-use-pep517 --no-deps --disable-pip-version-check -e .
+pytest -nauto tests --doctest-modules src/
+
+# Check that the readme example will work by running via doctest.
+# We run outside the repo to make the test a bit more similar to
+# a user running after installing with conda.
+mkdir ../temp
+cp README.md ../temp
+cd ../temp
+python -m doctest -v README.md

CHANGELOG.rst

@@ -14,6 +14,10 @@ Changelog
 
 - The :meth:`score` method of :class:`~glum.GeneralizedLinearRegressor` and :class:`~glum.GeneralizedLinearRegressorCV` now accepts offsets.
 
+**Other:**
+
+- The CI now runs daily unit tests against the nightly builds of numpy, pandas and scikit-learn.
+
 2.0.3 - 2021-11-05
 ------------------
 

src/glum/_glm.py

@@ -1079,7 +1079,7 @@ def report_diagnostics(
         import pandas as pd
 
         print("Diagnostics:")
-        with pd.option_context("max_rows", None, "max_columns", None):
+        with pd.option_context("display.max_rows", None, "display.max_columns", None):
             print(diagnostics)
 
     def get_formatted_diagnostics(

src/glum_benchmarks/bench_glum.py

@@ -95,7 +95,12 @@ def glum_bench(
         result["best_alpha"] = m.alpha_
 
     with pd.option_context(
-        "display.expand_frame_repr", False, "max_columns", None, "max_rows", None
+        "display.expand_frame_repr",
+        False,
+        "display.max_columns",
+        None,
+        "display.max_rows",
+        None,
     ):
         if diagnostics_level == "basic":
             m.report_diagnostics()

src/glum_benchmarks/cli_analyze.py

@@ -84,7 +84,12 @@ def cli_analyze(
     )
 
     with pd.option_context(
-        "display.expand_frame_repr", False, "max_columns", None, "max_rows", None
+        "display.expand_frame_repr",
+        False,
+        "display.max_columns",
+        None,
+        "display.max_rows",
+        None,
     ):
         if cols is not None:
             cols_to_show = _get_comma_sep_names(cols)