Manage coord values as tuples

And always make them numpy arrays for the InferenceData conversion.

Closes #5043

Author: michaelosthege

pymc/backends/arviz.py

@@ -222,10 +222,12 @@ def arbitrary_element(dct: Dict[Any, np.ndarray]) -> np.ndarray:
             aelem = arbitrary_element(get_from)
             self.ndraws = aelem.shape[0]
 
-        self.coords = {} if coords is None else coords
-        if hasattr(self.model, "coords"):
-            self.coords = {**self.model.coords, **self.coords}
-        self.coords = {key: value for key, value in self.coords.items() if value is not None}
+        self.coords = {**self.model.coords, **(coords or {})}
+        self.coords = {
+            cname: np.array(cvals) if isinstance(cvals, tuple) else cvals
+            for cname, cvals in self.coords.items()
+            if cvals is not None
+        }
 
         self.dims = {} if dims is None else dims
         if hasattr(self.model, "RV_dims"):

pymc/model.py

@@ -871,7 +871,7 @@ def RV_dims(self) -> Dict[str, Tuple[Union[str, None], ...]]:
         return self._RV_dims
 
     @property
-    def coords(self) -> Dict[str, Union[Sequence, None]]:
+    def coords(self) -> Dict[str, Union[Tuple, None]]:
         """Coordinate values for model dimensions."""
         return self._coords
 
@@ -1096,8 +1096,12 @@ def add_coord(
             raise ValueError(
                 f"The `length` passed for the '{name}' coord must be an Aesara Variable or None."
             )
+        if values is not None:
+            # Conversion to a tuple ensures that the coordinate values are immutable.
+            # Also unlike numpy arrays the's tuple.index(...) which is handy to work with.
+            values = tuple(values)
         if name in self.coords:
-            if not values.equals(self.coords[name]):
+            if not np.array_equal(values, self.coords[name]):
                 raise ValueError(f"Duplicate and incompatible coordinate: {name}.")
         else:
             self._coords[name] = values

pymc/tests/test_data_container.py

@@ -287,12 +287,12 @@ def test_explicit_coords(self):
             pm.Data("observations", data, dims=("rows", "columns"))
 
         assert "rows" in pmodel.coords
-        assert pmodel.coords["rows"] == ["R1", "R2", "R3", "R4", "R5"]
+        assert pmodel.coords["rows"] == ("R1", "R2", "R3", "R4", "R5")
         assert "rows" in pmodel.dim_lengths
         assert isinstance(pmodel.dim_lengths["rows"], ScalarSharedVariable)
         assert pmodel.dim_lengths["rows"].eval() == 5
         assert "columns" in pmodel.coords
-        assert pmodel.coords["columns"] == ["C1", "C2", "C3", "C4", "C5", "C6", "C7"]
+        assert pmodel.coords["columns"] == ("C1", "C2", "C3", "C4", "C5", "C6", "C7")
         assert pmodel.RV_dims == {"observations": ("rows", "columns")}
         assert "columns" in pmodel.dim_lengths
         assert isinstance(pmodel.dim_lengths["columns"], ScalarSharedVariable)

pymc/tests/test_idata_conversion.py

@@ -604,9 +604,10 @@ def test_constant_data_coords_issue_5046(self):
 
     def test_issue_5043_autoconvert_coord_values(self):
         coords = {
-            "city": ("Bonn", "Berlin"),
+            "city": pd.Series(["Bonn", "Berlin"]),
         }
         with pm.Model(coords=coords) as pmodel:
+            # The model tracks coord values as (immutable) tuples
             assert isinstance(pmodel.coords["city"], tuple)
             pm.Normal("x", dims="city")
             mtrace = pm.sample(
@@ -617,9 +618,28 @@ def test_issue_5043_autoconvert_coord_values(self):
                 tune=7,
                 draws=15,
             )
+            # The converter must convert coord values them to numpy arrays
+            # because tuples as coordinate values causes problems with xarray.
             converter = InferenceDataConverter(trace=mtrace)
-            with pytest.raises(ValueError, match="same length as the number of data dimensions"):
-                converter.to_inference_data()
+            assert isinstance(converter.coords["city"], np.ndarray)
+            converter.to_inference_data()
+
+            # We're not automatically converting things other than tuple,
+            # so advanced use cases remain supported at the InferenceData level.
+            # They just can't be used in the model construction already.
+            converter = InferenceDataConverter(
+                trace=mtrace,
+                coords={
+                    "city": pd.MultiIndex.from_tuples(
+                        [
+                            ("Bonn", 53111),
+                            ("Berlin", 10178),
+                        ],
+                        names=["name", "zipcode"],
+                    )
+                },
+            )
+            assert isinstance(converter.coords["city"], pd.MultiIndex)
 
 
 class TestPyMCWarmupHandling: