More Python tests for wrong ref in hypervolume and negative input in epsilon_mult().

 * python/src/moocore/_utils.py (array_1d_of_length_n): Add 'name' argument.
 * python/src/moocore/_moocore.py: Use it everywhere. Delete unreachable
   condition.
 * python/tests/test_moocore.py (test_wrong_ref): Add more tests.
   (check_roll_column, test_epsilon_mult_negative_input): New.

Author: MLopez-Ibanez

python/src/moocore/_moocore.py

@@ -154,7 +154,7 @@ def read_datasets(filename: str | os.PathLike | StringIO) -> np.ndarray:
 
 def _parse_maximise(maximise: bool | Sequence[bool], nobj: int) -> np.ndarray:
     """Convert maximise array or single bool to ndarray format."""
-    return array_1d_of_length_n(maximise, nobj).astype(bool)
+    return array_1d_of_length_n(maximise, nobj, name="maximise").astype(bool)
 
 
 def _parse_maximise_to_bool_array(
@@ -532,14 +532,10 @@ def hypervolume(
     # an int array.
     data, data_copied = asarray_maybe_copy(data)
     nobj = data.shape[1]
-    # Make sure it is a 1D array of length nobj.
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
-    if nobj != ref.shape[0]:
-        raise ValueError(
-            f"data and ref need to have the same number of objectives ({nobj} != {ref.shape[0]})"
-        )
     if nobj == 0:
         raise ValueError("input data must have at least 1 column")
+    # Make sure it is a 1D array of length nobj.
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
 
     maximise = _parse_maximise(maximise, nobj)
     # FIXME: Do this in C.
@@ -826,12 +822,7 @@ def hv_contributions(
     # an int array.
     x, x_copied = asarray_maybe_copy(x)
     nobj = x.shape[1]
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
-    if nobj != ref.shape[0]:
-        raise ValueError(
-            f"data and ref need to have the same number of objectives ({nobj} != {ref.shape[0]})"
-        )
-
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
     maximise = _parse_maximise(maximise, nobj)
     # FIXME: Do this in C.
     if maximise.any():
@@ -1006,11 +997,7 @@ def hv_approx(
     # an int array.
     data = np.asarray(data, dtype=float)
     nobj = data.shape[1]
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
-    if nobj != ref.shape[0]:
-        raise ValueError(
-            f"data and ref need to have the same number of objectives ({nobj} != {ref.shape[0]})"
-        )
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
 
     if not is_integer_value(nsamples):
         raise ValueError(f"nsamples must be an integer value: {nsamples}")
@@ -1766,8 +1753,12 @@ def normalise(
     to_range = np.asarray(to_range, dtype=float)
     if to_range.shape[0] != 2:
         raise ValueError("'to_range' must have length 2")
-    lower = array_1d_of_length_n(np.asarray(lower, dtype=float), nobj)
-    upper = array_1d_of_length_n(np.asarray(upper, dtype=float), nobj)
+    lower = array_1d_of_length_n(
+        np.asarray(lower, dtype=float), nobj, name="lower"
+    )
+    upper = array_1d_of_length_n(
+        np.asarray(upper, dtype=float), nobj, name="upper"
+    )
     if np.any(np.isnan(lower)):
         lower = np.where(np.isnan(lower), data.min(axis=0), lower)
     if np.any(np.isnan(upper)):
@@ -2295,7 +2286,7 @@ def whv_rect(
     #     else:
     #         pos = np.flatnonzero(maximise) + [0,2]
     #         rectangles[:, pos] = -rectangles[:, pos]
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
     ref = ffi.from_buffer("double []", ref)
     x, npoints, _ = np2d_to_double_array(x)
     rectangles, rectangles_nrow, _ = np2d_to_double_array(rectangles)
@@ -2401,14 +2392,16 @@ def total_whv_rect(
     if scalefactor <= 0 or scalefactor > 1:
         raise ValueError("'scalefactor' must be within (0,1]")
 
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
     whv = whv_rect(x, rectangles, ref=ref, maximise=maximise)
     hv = hypervolume(x, ref=ref)  # FIXME: maximise = maximise)
     if ideal is None:
         # FIXME: Should we include the range of the rectangles here?
         ideal = get_ideal(x, maximise=maximise)
     else:
-        ideal = array_1d_of_length_n(np.asarray(ideal, dtype=float), nobj)
+        ideal = array_1d_of_length_n(
+            np.asarray(ideal, dtype=float), nobj, name="ideal"
+        )
 
     beta = scalefactor * abs((ref - ideal).prod())
     return float(hv + beta * whv)
@@ -2661,8 +2654,10 @@ def whv_hype(
     if nobj != 2:
         raise NotImplementedError("Only 2D datasets are currently supported")
 
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
-    ideal = array_1d_of_length_n(np.asarray(ideal, dtype=float), nobj)
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
+    ideal = array_1d_of_length_n(
+        np.asarray(ideal, dtype=float), nobj, name="ideal"
+    )
     maximise = _parse_maximise(maximise, nobj)
     # FIXME: Do this in C.
     if maximise.any():
@@ -2688,7 +2683,7 @@ def whv_hype(
         mu = ffi.cast("double", mu)
         hv = lib.whv_hype_expo(data_p, npoints, ideal, ref, nsamples, seed, mu)
     elif dist == "point":
-        mu = array_1d_of_length_n(np.asarray(mu, dtype=float), nobj)
+        mu = array_1d_of_length_n(np.asarray(mu, dtype=float), nobj, name="mu")
         mu, _ = np1d_to_double_array(mu)
         hv = lib.whv_hype_gaus(data_p, npoints, ideal, ref, nsamples, seed, mu)
     else:
@@ -2880,14 +2875,10 @@ def r2_exact(
     # an int array.
     data, data_copied = asarray_maybe_copy(data)
     nobj = data.shape[1]
-    # Make sure it is a 1D array of length nobj.
-    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj)
-    if nobj != ref.shape[0]:
-        raise ValueError(
-            f"data and ref need to have the same number of objectives ({nobj} != {ref.shape[0]})"
-        )
     if nobj != 2:
         raise NotImplementedError("Only 2D datasets are currently supported")
+    # Make sure it is a 1D array of length nobj.
+    ref = array_1d_of_length_n(np.asarray(ref, dtype=float), nobj, name="ref")
 
     maximise = _parse_maximise(maximise, nobj)
     # FIXME: Do this in C.

python/src/moocore/_utils.py

@@ -61,14 +61,14 @@ def np1d_to_int_array(
     return np1d_to_c_array(x, ctype_data="int", ctype_size=ctype_size)
 
 
-def array_1d_of_length_n(x: ArrayLike, n: int) -> np.ndarray:
+def array_1d_of_length_n(x: ArrayLike, n: int, name: str = "x") -> np.ndarray:
     x = np.ravel(x)
     if len(x) == 1:
         return np.full((n), x[0])
     if x.shape[0] == n:
         return x
     raise ValueError(
-        f"1D array must have length {n} but it has length {x.shape[0]}"
+        f"{name!r} must have length {n}, but it has length {x.shape[0]}"
     )
 
 

python/tests/test_moocore.py

@@ -100,22 +100,35 @@ def test_hv_output(self, test_datapath, immutable_call):
 
         ref = np.array([2, -2, 2], dtype=float)
         x = [[1, 0, 1], [0, -1, 0]]
+        maximise = [False, True, False]
         assert (
-            immutable_call(
-                moocore.hypervolume, x, ref, maximise=[False, True, False]
-            )
+            immutable_call(moocore.hypervolume, x, ref, maximise=maximise)
             == 5.0
         )
+        hv = moocore.Hypervolume(ref=ref, maximise=maximise)
+        assert immutable_call(hv, x) == 5.0
 
         ref = [-2, -2, -2]
         x = [[-1, 0, -1], [0, -1, 0]]
         assert immutable_call(moocore.hypervolume, x, ref, maximise=True) == 5.0
 
-    def test_hv_wrong_ref(self, test_datapath):
-        """Check that the moocore.hypervolume() fails correctly after a ref with the wrong dimensions is input."""
-        X = self.input1
-        with pytest.raises(ValueError):
-            moocore.hypervolume(X[X[:, 2] == 1, :2], ref=np.array([10, 10, 10]))
+
+def test_wrong_ref():
+    """Check that the moocore.hypervolume() fails correctly after a ref with the wrong dimensions is input."""
+    x = [[1, 1, 1]]
+    ref = [10, 10]
+    hv = moocore.Hypervolume(ref=ref)
+    with pytest.raises(ValueError, match="same number of"):
+        hv(x)
+
+    with pytest.raises(ValueError, match="must have length 3"):
+        moocore.hypervolume(x, ref=ref)
+
+    with pytest.raises(ValueError, match="must have length 3"):
+        moocore.hv_contributions(x, ref=ref)
+
+    with pytest.raises(ValueError, match="must have length 3"):
+        moocore.hv_approx(x, ref=ref)
 
 
 @pytest.mark.parametrize("dim", range(5, 9))
@@ -229,6 +242,7 @@ def test_igd():
 
 @pytest.mark.parametrize("dim", range(0, 3))
 def test_is_nondominated_keep_weakly(dim):
+
     def check_keep_weakly(x, true_ndom, true_wndom):
         true_ndom = np.array(true_ndom)
         true_wndom = np.array(true_wndom)
@@ -242,6 +256,17 @@ def check_keep_weakly(x, true_ndom, true_wndom):
         test_weak_filter = moocore.filter_dominated(x, keep_weakly=True)
         assert_array_equal(test_weak_filter, x[true_wndom, :])
 
+    def check_roll_column(x, dim):
+        true_ndom = moocore.is_nondominated(x)
+        true_wndom = moocore.is_nondominated(x, keep_weakly=True)
+        a = np.append(np.zeros((len(x), dim)), x, axis=1)
+        for i in range(a.shape[1]):
+            x = np.roll(a, i, axis=1)
+            assert_array_equal(true_ndom, moocore.is_nondominated(x))
+            assert_array_equal(
+                true_wndom, moocore.is_nondominated(x, keep_weakly=True)
+            )
+
     x = np.array(
         [[2, 0], [1, 1], [3, 0], [2, 0], [1, 2], [3, 0], [0, 2], [1, 1], [1, 1]]
     )
@@ -250,6 +275,7 @@ def check_keep_weakly(x, true_ndom, true_wndom):
         true_ndom=[True, True, False, False, False, False, True, False, False],
         true_wndom=[True, True, False, True, False, False, True, True, True],
     )
+    check_roll_column(np.vstack((x, x)), dim)
 
     x = np.array(
         [[1, 0, 1], [1, 1, 1], [0, 1, 1], [1, 0, 1], [1, 1, 0], [1, 1, 1]]
@@ -259,6 +285,8 @@ def check_keep_weakly(x, true_ndom, true_wndom):
         true_ndom=[True, False, True, False, True, False],
         true_wndom=[True, False, True, True, True, False],
     )
+    # KUNG_SMALL_THRESHOLD is 16, so use 18 points.
+    check_roll_column(np.vstack((x, x, x)), dim)
 
 
 def test_is_nondominated(test_datapath):
@@ -299,7 +327,7 @@ def test_is_nondominated(test_datapath):
     )
 
 
-def test_epsilon(immutable_call):
+def test_epsilon():
     """Same as in R package."""
     ref = np.array([10, 1, 6, 1, 2, 2, 1, 6, 1, 10]).reshape((-1, 2))
     A = np.array([4, 2, 3, 3, 2, 4]).reshape((-1, 2))
@@ -313,6 +341,15 @@ def test_epsilon(immutable_call):
     assert_expected(2.5, moocore.epsilon_mult, A, ref, maximise=[False, True])
 
 
+def test_epsilon_mult_negative_input():
+    x = np.array([[-1, 2]])
+    y = np.array([[1, 2]])
+    with pytest.raises(ValueError, match="larger than 0"):
+        moocore.epsilon_mult(x, ref=y)
+    with pytest.raises(ValueError, match="larger than 0"):
+        moocore.epsilon_mult(y, ref=x)
+
+
 @pytest.mark.parametrize("dim", range(3, 6))
 def test_epsilon_dim(dim):
     seed = np.random.default_rng().integers(2**32 - 2)