Upgrade to Python 3.13

.github/workflows/pr.yaml

@@ -34,15 +34,22 @@ jobs:
     strategy:
       matrix:
         os: [ ubuntu-latest, windows-latest ]
+        python-version: [ "3.12", "3.13" ]
       fail-fast: false
     runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout repo
         uses: actions/checkout@v3
-      - name: Set up Python
+      - name: Enable long paths in Git (Windows Only)
+        if: runner.os == 'Windows'
+        run: git config --system core.longpaths true
+      - name: Enable Win32 long paths via registry (Windows Only)
+        if: runner.os == 'Windows'
+        run: reg add "HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem" /v LongPathsEnabled /t REG_DWORD /d 1 /f
+      - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v4
         with:
-          python-version: "3.12"
+          python-version: ${{ matrix.python-version }}
       - name: Install package
         run: make install
       - name: Run tests

Makefile

@@ -10,7 +10,7 @@ format:
 
 install:
 	pip install -e ".[dev]" --config-settings editable_mode=compat
-	pip install policyengine-us
+	pip install git+https://github.com/noman404/policyengine-us.git@noman404/python3.13
 	pip install policyengine-uk
 
 test-country-template:

changelog_entry.yaml

@@ -0,0 +1,5 @@
+- bump: major
+  changes:
+    changed:
+    - python 3.13.0
+    - numpy 2.1.0

policyengine_core/commons/formulas.py

@@ -92,6 +92,11 @@ def concat(this: ArrayLike[str], that: ArrayLike[str]) -> ArrayType[str]:
         array(['this1.0', 'that2.5']...)
 
     """
+    if isinstance(this, tuple):
+        raise TypeError("First argument must not be a tuple.")
+
+    if isinstance(that, tuple):
+        raise TypeError("Second argument must not be a tuple.")
 
     if isinstance(this, numpy.ndarray) and not numpy.issubdtype(
         this.dtype, numpy.str_

policyengine_core/enums/enum_array.py

@@ -96,7 +96,8 @@ def decode_to_str(self) -> numpy.str_:
         """
         return numpy.select(
             [self == item.index for item in self.possible_values],
-            [item.name for item in self.possible_values],
+            [str(item.name) for item in self.possible_values],
+            default="unknown",
         )
 
     def __repr__(self) -> str:

policyengine_core/parameters/vectorial_parameter_node_at_instant.py

@@ -203,13 +203,15 @@ def __getitem__(self, key: str) -> Any:
                     enum = type(key[0])
                     key = numpy.select(
                         [key == item for item in enum],
-                        [item.name for item in enum],
+                        [str(item.name) for item in enum],
+                        default="unknown",
                     )
                 elif isinstance(key, EnumArray):
                     enum = key.possible_values
                     key = numpy.select(
                         [key == item.index for item in enum],
                         [item.name for item in enum],
+                        default="unknown",
                     )
                 else:
                     key = key.astype("str")

policyengine_core/populations/group_population.py

@@ -232,7 +232,7 @@ def max(self, array: ArrayLike, role: Role = None) -> ArrayLike:
         return self.reduce(
             array,
             reducer=numpy.maximum,
-            neutral_element=-numpy.infty,
+            neutral_element=-numpy.inf,
             role=role,
         )
 
@@ -256,7 +256,7 @@ def min(self, array: ArrayLike, role: Role = None) -> ArrayLike:
         return self.reduce(
             array,
             reducer=numpy.minimum,
-            neutral_element=numpy.infty,
+            neutral_element=numpy.inf,
             role=role,
         )
 

policyengine_core/populations/population.py

@@ -253,7 +253,11 @@ def get_rank(
         # We double-argsort all lines of the matrix.
         # Double-argsorting gets the rank of each value once sorted
         # For instance, if x = [3,1,6,4,0], y =  numpy.argsort(x) is [4, 1, 0, 3, 2] (because the value with index 4 is the smallest one, the value with index 1 the second smallest, etc.) and z =  numpy.argsort(y) is [2, 1, 4, 3, 0], the rank of each value.
-        sorted_matrix = numpy.argsort(numpy.argsort(matrix))
+
+        # because of the infinities the first sort creates positional indices
+        # The second argsort converts these positions to ranks, thus fixes the broken sort issue
+        first_argsort = numpy.argsort(matrix, axis=1, kind="stable")
+        sorted_matrix = numpy.argsort(first_argsort, axis=1, kind="stable")
 
         # Build the result vector by taking for each person the value in the right line (corresponding to its household id) and the right column (corresponding to its position)
         result = sorted_matrix[ids, positions]

policyengine_core/taxscales/marginal_rate_tax_scale.py

@@ -66,12 +66,12 @@ def calc(
         #
         #   numpy.finfo(float_).eps
         thresholds1 = numpy.outer(
-            factor + numpy.finfo(numpy.float_).eps,
+            factor + numpy.finfo(numpy.float64).eps,
             numpy.array(self.thresholds + [numpy.inf]),
         )
 
         if round_base_decimals is not None:
-            thresholds1 = numpy.round_(thresholds1, round_base_decimals)
+            thresholds1 = numpy.round(thresholds1, round_base_decimals)
 
         a = numpy.maximum(
             numpy.minimum(base1, thresholds1[:, 1:]) - thresholds1[:, :-1], 0
@@ -82,8 +82,8 @@ def calc(
 
         else:
             r = numpy.tile(self.rates, (len(tax_base), 1))
-            b = numpy.round_(a, round_base_decimals)
-            return numpy.round_(r * b, round_base_decimals).sum(axis=1)
+            b = numpy.round(a, round_base_decimals)
+            return numpy.round(r * b, round_base_decimals).sum(axis=1)
 
     def combine_bracket(
         self,

policyengine_core/taxscales/rate_tax_scale_like.py

@@ -175,12 +175,12 @@ def bracket_indices(
         #
         #   numpy.finfo(float_).eps
         thresholds1 = numpy.outer(
-            +factor + numpy.finfo(numpy.float_).eps,
+            +factor + numpy.finfo(numpy.float64).eps,
             numpy.array(self.thresholds),
         )
 
         if round_decimals is not None:
-            thresholds1 = numpy.round_(thresholds1, round_decimals)
+            thresholds1 = numpy.round(thresholds1, round_decimals)
 
         return (base1 - thresholds1 >= 0).sum(axis=1) - 1
 

policyengine_core/tools/simulation_dumper.py

@@ -95,7 +95,8 @@ def _dump_entity(population, directory):
     else:
         encoded_roles = np.select(
             [population.members_role == role for role in flattened_roles],
-            [role.key for role in flattened_roles],
+            [str(role.key) for role in flattened_roles],
+            default="unknown",
         )
     np.save(os.path.join(path, "members_role.npy"), encoded_roles)
 

setup.py

@@ -11,7 +11,7 @@
 
 general_requirements = [
     "pytest>=8,<9",
-    "numpy~=1.26.4",
+    "numpy~=2.1.0",
     "sortedcontainers<3",
     "numexpr<3",
     "dpath<3",
@@ -25,6 +25,7 @@
     "pyvis>=0.3.2",
     "microdf_python>=0.4.3",
     "huggingface_hub>=0.25.1",
+    "standard-imghdr",
 ]
 
 dev_requirements = [
@@ -60,6 +61,7 @@
         "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
         "Programming Language :: Python :: 3.12",
+        "Programming Language :: Python :: 3.13",
         "Topic :: Scientific/Engineering :: Information Analysis",
     ],
     description="Core microsimulation engine enabling country-specific policy models.",

tests/core/commons/test_formulas.py

@@ -79,3 +79,13 @@ def test_switch_when_values_are_empty():
 
     with pytest.raises(AssertionError):
         assert commons.switch(conditions, value_by_condition)
+
+
+def test_concat_tuple_inputs():
+    with pytest.raises(TypeError, match="First argument must not be a tuple."):
+        commons.concat(("a", "b"), numpy.array(["c", "d"]))
+
+    with pytest.raises(
+        TypeError, match="Second argument must not be a tuple."
+    ):
+        commons.concat(numpy.array(["a", "b"]), ("c", "d"))

tests/core/parameters_fancy_indexing/test_fancy_indexing.py

@@ -114,5 +114,10 @@ class TypesZone(Enum):
         z1 = "Zone 1"
         z2 = "Zone 2"
 
-    zone = np.asarray([TypesZone.z1, TypesZone.z2, TypesZone.z2, TypesZone.z1])
+    zone = np.asarray(
+        [
+            z.name
+            for z in [TypesZone.z1, TypesZone.z2, TypesZone.z2, TypesZone.z1]
+        ]
+    )
     assert_near(P.single.owner[zone], [100, 200, 200, 100])

tests/core/test_tracers.py

@@ -403,7 +403,8 @@ def test_log_aggregate(tracer):
 
     lines = tracer.computation_log.lines(aggregate=True)
     assert (
-        lines[0] == "  A<2017, (default)> = {'avg': 1.0, 'max': 1, 'min': 1}"
+        lines[0].strip()
+        == "A<2017, (default)> = {'avg': np.float64(1.0), 'max': np.int64(1), 'min': np.int64(1)}"
     )