fix: validate convolution filter dimensions and unique axes

Author: antonsoo

pylops/signalprocessing/convolvend.py

@@ -64,6 +64,12 @@ class ConvolveND(LinearOperator):
     shape : :obj:`tuple`
         Operator shape.
 
+    Raises
+    ------
+    ValueError
+        If ``h`` has more dimensions than ``dims``, or ``axes`` does not
+        contain one unique model axis per filter dimension.
+
     Notes
     -----
     The ConvolveND operator applies n-dimensional convolution
@@ -85,6 +91,9 @@ def __init__(
         name: str = "C",
     ):
         dims = _value_or_sized_to_tuple(dims)
+        if h.ndim > len(dims):
+            msg = "h cannot have more dimensions than dims"
+            raise ValueError(msg)
         super().__init__(dtype=np.dtype(dtype), dims=dims, dimsd=dims, name=name)
 
         self.axes = (
@@ -94,6 +103,12 @@ def __init__(
                 [get_normalize_axis_index()(ax, len(self.dims)) for ax in axes]
             )
         )
+        if len(self.axes) != h.ndim:
+            msg = "axes must contain one axis per dimension of h"
+            raise ValueError(msg)
+        if len(set(self.axes)) != len(self.axes):
+            msg = "axes must be unique"
+            raise ValueError(msg)
         self.h = h
         hshape = np.array(self.h.shape)
 

pytests/test_convolve.py

@@ -20,6 +20,8 @@
 from pylops.signalprocessing import Convolve1D, Convolve2D, ConvolveND
 from pylops.utils import dottest
 
+np.random.seed(10)
+
 # filters
 nfilt = (5, 6, 5)
 h1 = triang(nfilt[0], sym=True)
@@ -195,6 +197,43 @@ def test_Convolve1D_method_error():
         )
 
 
+@pytest.mark.parametrize("par", [(par1_3d), (par2_3d)])
+@pytest.mark.parametrize("axes", [(0, 1), (0, 1, 2, 0)])
+def test_ConvolveND_axes_length_error(par, axes):
+    """Error raised when axes do not match the filter dimensionality."""
+    with pytest.raises(ValueError, match="axes must contain one axis per dimension of h"):
+        ConvolveND(
+            dims=(par["nz"], par["ny"], par["nx"]),
+            h=np.random.normal(size=nfilt),
+            offset=par["offset"],
+            axes=axes,
+        )
+
+
+@pytest.mark.parametrize("par", [(par1_3d), (par2_3d)])
+@pytest.mark.parametrize("axes", [(0, 0, 2), (0, 1, -2)])
+def test_ConvolveND_axes_unique_error(par, axes):
+    """Error raised for repeated axes, including negative aliases."""
+    with pytest.raises(ValueError, match="axes must be unique"):
+        ConvolveND(
+            dims=(par["nz"], par["ny"], par["nx"]),
+            h=np.random.normal(size=nfilt),
+            offset=par["offset"],
+            axes=axes,
+        )
+
+
+@pytest.mark.parametrize("par", [(par1_3d), (par2_3d)])
+@pytest.mark.parametrize("hshape", [nfilt, (1, 1, 1)])
+def test_ConvolveND_shape_error(par, hshape):
+    """Error raised when the filter has more dimensions than the model."""
+    with pytest.raises(ValueError, match="h cannot have more dimensions than dims"):
+        ConvolveND(
+            dims=(par["ny"], par["nx"]),
+            h=np.random.normal(size=hshape),
+        )
+
+
 @pytest.mark.parametrize("dims", [nfilt[0] * 4, (3, nfilt[0] * 4)])
 @pytest.mark.parametrize("hlen", ["short", "long"])
 def test_Convolve1D_hstar_contiguous(dims, hlen):