Merge pull request #16 from 3DFin/develop_v0_3_0

v0.3.0

Author: drprojects

.github/workflows/wheels.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-22.04, windows-2022, macos-14]
+        os: [ubuntu-latest, windows-latest, macos-13, macos-latest]
       fail-fast: false
 
     steps:
@@ -27,14 +27,14 @@ jobs:
       - uses: actions/setup-python@v5
 
       - name: Install cibuildwheel
-        run: python -m pip install cibuildwheel==2.15
+        run: python -m pip install cibuildwheel==2.20.0
 
       - name: Build wheels
         run: python -m cibuildwheel --output-dir wheelhouse
 
-      - uses: actions/upload-artifact@v3
+      - uses: actions/upload-artifact@v4
         with:
-          name: wheels
+          name: cibw-wheels-${{ matrix.os }}-${{ strategy.job-index }}
           path: ./wheelhouse/*.whl
 
   publish:
@@ -48,20 +48,20 @@ jobs:
         with:
           submodules: True
 
-      - uses: actions/setup-python@v4
+      - uses: actions/setup-python@v5
 
-      - uses: actions/download-artifact@v3
+      - uses: actions/download-artifact@v4
         with:
-          name: wheels
-          path: wheelhouse
+          pattern: cibw-*
+          path: dist
+          merge-multiple: true
 
       - name: Publish to PyPI
         env:
           TWINE_USERNAME: __token__
           TWINE_PASSWORD: ${{ secrets.PYPI_TOKEN }}
         # construct the source package and upload src and wheels to PiPy
         run: |
-          python -m pip install twine
-          python setup.py sdist 
+          python -m pip install twine build --upgrade
+          python -m build --sdist
           twine upload dist/*
-          twine upload wheelhouse/*.whl

CMakeLists.txt

@@ -16,18 +16,19 @@ find_package(nanobind CONFIG REQUIRED)
 
 nanobind_add_module(pgeof_ext NOMINSIZE STABLE_ABI LTO src/pgeof_ext.cpp)
 
-# for nanobind 2
-# nanobind_add_stub(
-#   pgeof_ext_stub
-#   MODULE pgeof_ext
-#   OUTPUT pgeof_ext.pyi
-#   PYTHON_PATH $<TARGET_FILE_DIR:pgeof_ext>
-#   DEPENDS pgeof_ext
-# )
+nanobind_add_stub(
+   pgeof_ext_stub
+   MODULE pgeof_ext
+   OUTPUT pgeof_ext.pyi
+   MARKER_FILE py.typed
+   PYTHON_PATH $<TARGET_FILE_DIR:pgeof_ext>
+   DEPENDS pgeof_ext
+)
 
 # All lib are headeer only. 
 # it's faster to include like this than using exported targets 
 # (i.e add_subdirectories(...))
 target_include_directories(pgeof_ext PRIVATE "include" "third_party/eigen" "third_party/nanoflann/include" "third_party/taskflow") 
 
-install(TARGETS pgeof_ext LIBRARY DESTINATION pgeof)
+install(TARGETS pgeof_ext LIBRARY DESTINATION pgeof)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/pgeof_ext.pyi ${CMAKE_CURRENT_BINARY_DIR}/py.typed DESTINATION pgeof)

include/nn_search.hpp

@@ -36,7 +36,7 @@ static std::pair<nb::ndarray<nb::numpy, uint32_t, nb::ndim<2>>, nb::ndarray<nb::
 
     if (knn > data.rows()) { throw std::invalid_argument("knn size is greater than the data point cloud size"); }
 
-    kd_tree_t          kd_tree(3, data, 10);
+    kd_tree_t          kd_tree(3, data, 10, 0);
     const Eigen::Index n_points = query.rows();
     uint32_t*          indices  = new uint32_t[knn * n_points];
     nb::capsule        owner_indices(indices, [](void* p) noexcept { delete[] (uint32_t*)p; });
@@ -94,7 +94,7 @@ static std::pair<nb::ndarray<nb::numpy, int32_t, nb::ndim<2>>, nb::ndarray<nb::n
         throw std::invalid_argument("max knn size is greater than the data point cloud size");
     }
 
-    kd_tree_t    kd_tree(3, data, 10);
+    kd_tree_t    kd_tree(3, data, 10, 0);
     const real_t sq_search_radius = search_radius * search_radius;
 
     const Eigen::Index n_points = query.rows();

include/pgeof.hpp

@@ -73,9 +73,10 @@ static inline void flush() { std::cout << std::endl; };
  * @return the geometric features associated with each point's neighborhood in a (num_points, features_count) nd::array.
  */
 template <typename real_t = float, const size_t feature_count = 11>
-static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>> compute_geometric_features(
-    RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn, nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr,
-    const size_t k_min, const bool verbose)
+static nb::ndarray<nb::numpy, real_t, nb::shape<-1, static_cast<nb::ssize_t>(feature_count)>>
+    compute_geometric_features(
+        RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn,
+        nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr, const size_t k_min, const bool verbose)
 {
     if (k_min < 1) { throw std::invalid_argument("k_min should be > 1"); }
     // Each point can be treated in parallel
@@ -111,7 +112,8 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>> compute
     // Final print to start on a new line
     if (verbose) log::flush();
     const size_t shape[2] = {n_points, feature_count};
-    return nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>>(features, 2, shape, owner_features);
+    return nb::ndarray<nb::numpy, real_t, nb::shape<-1, static_cast<nb::ssize_t>(feature_count)>>(
+        features, 2, shape, owner_features);
 }
 /**
  * Convenience function that check that scales are well ordered in increasing order.
@@ -155,9 +157,10 @@ static bool check_scales(const std::vector<uint32_t>& k_scales)
  * nd::array
  */
 template <typename real_t, const size_t feature_count = 11>
-static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any, feature_count>> compute_geometric_features_multiscale(
-    RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn, nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr,
-    const std::vector<uint32_t>& k_scales, const bool verbose)
+static nb::ndarray<nb::numpy, real_t, nb::shape<-1, -1, static_cast<nb::ssize_t>(feature_count)>>
+    compute_geometric_features_multiscale(
+        RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn,
+        nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr, const std::vector<uint32_t>& k_scales, const bool verbose)
 {
     if (!check_scales(k_scales))
     {
@@ -203,7 +206,7 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any, feature_count>
     if (verbose) log::flush();
 
     const size_t shape[3] = {n_points, n_scales, feature_count};
-    return nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any, feature_count>>(
+    return nb::ndarray<nb::numpy, real_t, nb::shape<-1, -1, static_cast<nb::ssize_t>(feature_count)>>(
         features, 3, shape, owner_features);
 }
 
@@ -238,9 +241,11 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any, feature_count>
  * @return Geometric features associated with each point's neighborhood in a (num_points, features_count) nd::array
  */
 template <typename real_t, const size_t feature_count = 12>
-static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>> compute_geometric_features_optimal(
-    RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn, nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr,
-    const uint32_t k_min, const uint32_t k_step, const uint32_t k_min_search, const bool verbose)
+static nb::ndarray<nb::numpy, real_t, nb::shape<-1, static_cast<nb::ssize_t>(feature_count)>>
+    compute_geometric_features_optimal(
+        RefCloud<real_t> xyz, nb::ndarray<const uint32_t, nb::ndim<1>> nn,
+        nb::ndarray<const uint32_t, nb::ndim<1>> nn_ptr, const uint32_t k_min, const uint32_t k_step,
+        const uint32_t k_min_search, const bool verbose)
 {
     if (k_min < 1 && k_min_search < 1) { throw std::invalid_argument("k_min and k_min_search should be > 1"); }
     // Each point can be treated in parallel
@@ -300,7 +305,8 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>> compute
     if (verbose) log::flush();
 
     const size_t shape[2] = {n_points, feature_count};
-    return nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>>(features, 2, shape, owner_features);
+    return nb::ndarray<nb::numpy, real_t, nb::shape<-1, static_cast<nb::ssize_t>(feature_count)>>(
+        features, 2, shape, owner_features);
 }
 
 /**
@@ -317,14 +323,14 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, feature_count>> compute
  * @return Geometric features associated with each point's neighborhood in a (num_points, features_count) nd::array
  */
 template <typename real_t>
-static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any>> compute_geometric_features_selected(
+static nb::ndarray<nb::numpy, real_t, nb::shape<-1, -1>> compute_geometric_features_selected(
     RefCloud<real_t> xyz, const real_t search_radius, const uint32_t max_knn,
     const std::vector<EFeatureID>& selected_features)
 {
     using kd_tree_t = nanoflann::KDTreeEigenMatrixAdaptor<RefCloud<real_t>, 3, nanoflann::metric_L2_Simple>;
     // TODO: where knn < num of points
 
-    kd_tree_t          kd_tree(3, xyz, 10);
+    kd_tree_t          kd_tree(3, xyz, 10, 0);
     const size_t       feature_count    = selected_features.size();
     const Eigen::Index n_points         = xyz.rows();
     real_t             sq_search_radius = search_radius * search_radius;
@@ -364,6 +370,7 @@ static nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any>> compute_geome
         });
     executor.run(taskflow).get();
 
-    return nb::ndarray<nb::numpy, real_t, nb::shape<nb::any, nb::any>>(features, {static_cast<size_t>(n_points), feature_count}, owner_features);
+    return nb::ndarray<nb::numpy, real_t, nb::shape<-1, -1>>(
+        features, {static_cast<size_t>(n_points), feature_count}, owner_features);
 }
 }  // namespace pgeof

pyproject.toml

@@ -1,10 +1,13 @@
 [build-system]
-requires = ["scikit-build-core >=0.4.3", "nanobind == 1.9.2"]
+requires = ["scikit-build-core >=0.4.3", 
+  "nanobind == 2.1.0",
+  "typing_extensions;python_version < '3.11'"
+]
 build-backend = "scikit_build_core.build"
 
 [project]
 name = "pgeof"
-version = "0.2.0"
+version = "0.3.0"
 readme = "README.md"
 description = "Compute the geometric features associated with each point's neighborhood:"
 requires-python = ">=3.8,<3.13"
@@ -36,6 +39,47 @@ build-dir = "build/{wheel_tag}"
 
 cmake.build-type = "Release"
 
+# make sdist a lot lighter by removing some useless files from third_party
+# ⚠️  be sure to keep copyrights and license file
+sdist.exclude = [
+  "third_party/eigen/bench",
+  "third_party/eigen/demos",
+  "third_party/eigen/doc",
+  "third_party/taskflow/3rd-party",
+  "third_party/taskflow/benchmarks",
+  "third_party/taskflow/docs",
+  "third_party/taskflow/doxygen",
+  "third_party/taskflow/examples",
+  "third_party/taskflow/sandbox",
+  "third_party/taskflow/unittests",
+]
+
+[tool.ruff]
+target-version = "py310"
+line-length = 120
+
+[tool.ruff.lint]
+# TODO Add D, PTH, RET, disabled for now as they collides with intial choices
+select = ["E", "W", "YTT", "NPY", "PYI", "Q", "F", "B", "I", "SIM", "RUF"]
+# TODO: for now we ignore "Line too long error (E501)" 
+# because our comments are too longs
+# code formatting will take care of the line length in code anyway
+ignore = [
+  "E501",
+  # Ignore docstring in public package and module
+  "D100",
+  "D104",
+  # Blank line before class
+  "D203",
+  # multiline summary second line
+  "D213",
+  # yoda conditions
+  "SIM300",
+]
+
+[tool.ruff.lint.isort]
+known-first-party = ["pgeof"]
+
 [tool.tox]
 legacy_tox_ini = """
 [tox]
@@ -67,4 +111,4 @@ archs = ["auto64"] # limit to 64bits builds
 
 # Needed for full C++17 support
 [tool.cibuildwheel.macos.environment]
-MACOSX_DEPLOYMENT_TARGET = "10.14"
+MACOSX_DEPLOYMENT_TARGET = "11.0"

src/pgeof_ext.cpp

@@ -1,7 +1,6 @@
 
 #include <nanobind/nanobind.h>
 #include <nanobind/ndarray.h>
-#include <nanobind/stl/pair.h>
 #include <nanobind/stl/vector.h>
 
 #include "nn_search.hpp"
@@ -176,4 +175,4 @@ NB_MODULE(pgeof_ext, m)
             :param selected_features: List of selected features. See EFeatureID
             :return: Geometric features associated with each point's neighborhood in a (num_points, features_count) numpy array.
         )");
-}
+}

tests/bench_jakteristics.py

@@ -11,7 +11,8 @@
 
 @pytest.fixture
 def random_point_cloud():
-    return np.random.rand(1000000, 3) * 100
+    rng = np.random.default_rng()
+    return rng.random(0, 100, size=((1000000, 3)))
 
 
 @pytest.mark.benchmark(group="feature-computation-jak", disable_gc=True, warmup=True)

tests/bench_knn.py

@@ -7,7 +7,8 @@
 
 @pytest.fixture
 def random_point_cloud():
-    return np.random.rand(100000, 3).astype("float32")
+    rng = np.random.default_rng()
+    return rng.random(0, 100, size=((1000000, 3))).astype("float32")
 
 
 @pytest.mark.benchmark(group="knn", disable_gc=True, warmup=True)

tests/helpers.py

@@ -4,7 +4,8 @@
 
 def random_nn(num_points, k):
     # Generate a random synthetic point cloud
-    xyz = np.random.rand(num_points, 3)
+    rng = np.random.default_rng()
+    xyz = rng.random(size=(num_points, 3))
 
     # Converting k-nearest neighbors to CSR format
     kneigh = KDTree(xyz).query(xyz, k=k, workers=-1)
@@ -20,4 +21,4 @@ def random_nn(num_points, k):
     xyz = np.ascontiguousarray(xyz)
     nn_ptr = np.ascontiguousarray(nn_ptr)
     nn = np.ascontiguousarray(nn)
-    return xyz, nn, nn_ptr
+    return xyz, nn, nn_ptr

tests/test_pgeof.py

@@ -7,7 +7,8 @@
 
 def test_knn():
     knn = 10
-    xyz = np.random.rand(1000, 3)
+    rng = np.random.default_rng()
+    xyz = rng.random(size=(1000, 3))
     xyz = xyz.astype("float32")
     tree = KDTree(xyz)
     _, k_legacy = tree.query(xyz, k=knn, workers=-1)
@@ -18,7 +19,8 @@ def test_knn():
 def test_radius_search():
     knn = 10
     radius = 0.2
-    xyz = np.random.rand(1000, 3)
+    rng = np.random.default_rng()
+    xyz = rng.random(size=(1000, 3))
     xyz = xyz.astype("float32")
     tree = KDTree(xyz)
     _, k_legacy = tree.query(xyz, k=knn, distance_upper_bound=radius, workers=-1)
@@ -43,4 +45,4 @@ def test_pgeof_multiscale():
     simple = pgeof.compute_features(xyz, nn, nn_ptr, 50, False)
     multi_simple = pgeof.compute_features_multiscale(xyz, nn, nn_ptr, [20], False)
     np.testing.assert_allclose(multi[:, 0], multi_simple[:, 0], 1e-1, 1e-5)
-    np.testing.assert_allclose(multi[:, 1], simple, 1e-1, 1e-5)
+    np.testing.assert_allclose(multi[:, 1], simple, 1e-1, 1e-5)