v2.4.2

.github/workflows/python-test.yml

@@ -32,6 +32,7 @@ jobs:
     - name: Lint
       run: |
         uv run ruff check
+        uv run mypy .
     - name: Test
       run: |
         uv run pytest

humanleague/utils.py

@@ -18,7 +18,7 @@ def tabulate_counts(population: npt.NDArray, names: list[str] | tuple[str, ...]
         pd.Series: A pandas Series where the index is a MultiIndex created from the shape of the input array,
         and the data corresponds to the flattened values of the input array.
     """
-    index = pd.MultiIndex.from_tuples(list(np.ndindex(population.shape)), names=names)
+    index = pd.MultiIndex.from_tuples(list(np.ndindex(population.shape)), names=names)  # type: ignore[arg-type]
     return pd.Series(index=index, data=list(np.nditer(population)), dtype=int, name="count")
 
 

pyproject.toml

@@ -10,7 +10,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "humanleague"
-version = "2.4.1"
+version = "2.4.2"
 authors = [
   { name="Andrew Smith", email="andrew@friarswood.net" },
 ]
@@ -32,12 +32,13 @@ dependencies = [
 
 [dependency-groups]
 dev = [
-  "pybind11>=2.10.3",
+  "pybind11>=3.0.0",
   "pytest>=8.1.4",
   "mypy>=1.5.0",
   "mypy-extensions>=1.0.0",
-  "ruff>=0.0.286",
-  "build>=1.2.2.post1"
+  "ruff>=0.12.9",
+  "build>=1.2.2.post1",
+  "typing-extensions>=4.15.0",
 ]
 
 [tool.pytest.ini_options]

src/Integerise.cpp

@@ -11,10 +11,13 @@
 
 namespace {
 
-int64_t checked_round(double x, double tol=1e-4) // loose tolerance ~1/4 mantissa precision
+ // loose tolerance ~1/4 mantissa precision
+constexpr double TOL = 1e-4;
+
+int64_t checked_round(double x)
 {
-  if (fabs(x - round(x)) > tol)
-    throw std::runtime_error("Marginal or total value %% is not an integer (within tolerance %%)"s % x % tol);
+  if (fabs(x - round(x)) > TOL)
+    throw std::runtime_error("Marginal or total value %% is not an integer (within tolerance %%)"s % x % TOL);
   return (int64_t)round(x);
 }
 
@@ -62,7 +65,8 @@ Integeriser::Integeriser(const NDArray<double>& seed) : m_seed(seed)
   {
     // convert to vector (reduce 1-d special case)
     std::vector<double> p = reduce(seed, 0);
-    int pop = sum(seed);
+    // casting rounds down so for better consistency (with checked_round) add TOL
+    int pop = sum(seed) + TOL;
     // convert to probabilities
     for (auto& x: p) x /= pop;
     std::vector<int> tmp = integeriseMarginalDistribution(p, pop, m_rmse);

src/module.cpp

@@ -130,6 +130,7 @@ py::tuple integerise1d(py::array_t<double> frac_a, int pop) {
   const std::vector<int>& freq = integeriseMarginalDistribution(prob, pop, var);
 
   py::dict stats;
+  stats["conv"] = true; // always converges, but including for consistency
   stats["rmse"] = var;
 
   return py::make_tuple(py::array_t<int>(freq.size(), freq.data()), stats);

tests/test_all.py

@@ -78,12 +78,14 @@ def test_integerise() -> None:
     # exact
     r, stats = hl.integerise(np.array([0.4, 0.3, 0.2, 0.1]), 10)
     assert stats["rmse"] < 1e-15
+    assert stats["conv"]  # 1d case with specified total will always converge, but return for consistency
     assert np.array_equal(r, np.array([4, 3, 2, 1]))
 
     # inexact
     r, stats = hl.integerise(np.array([0.4, 0.3, 0.2, 0.1]), 17)
     assert stats["rmse"] == pytest.approx(0.273861278752583, abs=1e-6)
-
+    # without total we still get the same stats keys (not values)
+    assert stats.keys() == hl.integerise(np.array([0.4, 0.3, 0.2, 0.1]))[1].keys()
     assert np.array_equal(r, np.array([7, 5, 3, 2]))
 
     # 1-d case
@@ -113,6 +115,26 @@ def test_integerise() -> None:
     assert np.sum(result) == sum(m0)
     assert stats["rmse"] < 1.05717
 
+    # 1d integerise without providing total - check total rounds up if appropriate
+    a = np.array([1.1, 2.9, 0.9999])
+    result, stats = hl.integerise(a)
+    assert (result == np.array([1, 3, 1])).all()
+    assert stats["conv"]
+
+    # 1d integerise without providing total - check total doesnt round up if inappropriate
+    a[2] = 1.0001
+    result, stats = hl.integerise(a)
+    assert (result == np.array([1, 3, 1])).all()
+    assert stats["conv"]
+
+    # outside tolerance
+    a[2] = 1.0002
+    with pytest.raises(RuntimeError):
+        hl.integerise(a)
+    a[2] = 0.9998
+    with pytest.raises(RuntimeError):
+        hl.integerise(a)
+
 
 def test_IPF() -> None:
     m0 = np.array([52.0, 48.0])

tests/test_utils.py

@@ -22,7 +22,7 @@ def test_tabulate_counts_no_names() -> None:
     population = np.array([[5, 6], [7, 8]])
     result = hl.tabulate_counts(population)
 
-    expected_index = pd.MultiIndex.from_tuples([(0, 0), (0, 1), (1, 0), (1, 1)], names=None)
+    expected_index = pd.MultiIndex.from_tuples([(0, 0), (0, 1), (1, 0), (1, 1)], names=None)  # type: ignore[arg-type]
     expected_data = [5, 6, 7, 8]
     expected = pd.Series(data=expected_data, index=expected_index, name="count")