deps: update numerical and contributor dependencies

.github/workflows/ci.yml

@@ -189,6 +189,11 @@ jobs:
         uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
         with:
           python-version: ${{ matrix.python-version }}
+      - name: Install macOS OpenMP runtime for optional ML
+        if: runner.os == 'macOS'
+        env:
+          HOMEBREW_NO_AUTO_UPDATE: "1"
+        run: brew install libomp
       - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
         with:
           name: release-candidate
@@ -201,7 +206,7 @@ jobs:
           --expected-tree "$(git rev-parse 'HEAD^{tree}')"
       - name: Export installed-wheel test environment
         run: >-
-          uv export --locked --group dev --extra ta --extra store --extra viz
+          uv export --locked --group dev --extra ta --extra store --extra viz --extra ml
           --no-emit-project --no-hashes
           --output-file "${{ runner.temp }}/test-requirements.txt"
       - name: Install the candidate wheel in a clean environment
@@ -217,6 +222,12 @@ jobs:
         run: >-
           uv run --python .artifact-venv --no-project
           python scripts/ci/candidate.py verify-installed candidate --source-root src
+      - name: Import optional ML dependencies
+        run: >-
+          uv run --python .artifact-venv --no-project
+          python -c "import numpy, scipy, sklearn, shap, lightgbm;
+          print(numpy.__version__, scipy.__version__, sklearn.__version__,
+          shap.__version__, lightgbm.__version__)"
       - name: Run ty check
         run: >-
           uv run --python .artifact-venv --no-project

docs/getting-started/installation.md

@@ -6,13 +6,13 @@ and then move directly into the first workflow.
 ## Requirements
 
 - Python 3.12, 3.13, 3.14, or 3.15
-- Polars 0.20+
+- Polars 2.0+
 
-These requirements describe development source for the 0.2 release. Published 0.1.6
+The published 0.2.0 release supports these Python versions and the `calendar` argument.
+Published 0.1.6
 supports Python 3.12-3.14 and does not accept the new `calendar` argument. Use its
 [version-matched quick start](https://github.com/ml4t/engineer/tree/v0.1.6#quick-start)
-for that release. Until 0.2 is available from PyPI, use the source installation below
-for Python 3.15 and the calendar-aware examples in this guide.
+for that release.
 
 ## Install from PyPI
 
@@ -73,6 +73,24 @@ pip install "ml4t-engineer[stats]"
 
 The core `ffdiff()` transform does not require this extra.
 
+### Optional ML dependencies
+
+The `ml` extra installs LightGBM and SHAP:
+
+```bash
+uv pip install "ml4t-engineer[ml]"
+```
+
+On macOS, LightGBM also needs the OpenMP runtime. Install it before importing the ML
+dependencies, including when using the `dev` or `all` extras:
+
+```bash
+brew install libomp
+```
+
+See the [LightGBM Python installation guide](https://github.com/lightgbm-org/LightGBM/blob/v4.7.0/python-package/README.rst#installation)
+for native dependency details.
+
 ## Verify Installation
 
 ```python
@@ -84,11 +102,7 @@ print(f"Total features: {len(all_features)}")
 print(f"Categories: {feature_catalog.categories()}")
 ```
 
-Expected output:
-```
-Total features: 120
-Categories: ['math', 'microstructure', 'ml', 'momentum', 'price_transform', 'regime', 'risk', 'statistics', 'trend', 'volatility', 'volume']
-```
+This prints the features and categories available in your installed release.
 
 ## Next Steps