Merge pull request #14794 from larrybradley/discretize-factor-int

Raise nicer error message if factor is not integer

astropy/convolution/tests/test_discretize.py

@@ -265,3 +265,10 @@ def test_discretize_oversample():
     assert_allclose(vmax, 0.927, atol=1e-3)
     assert vmax_yx == (25, 5)
     assert_allclose(values_center, values_osf1)
+
+
+def test_oversample_factor():
+    gauss_1D = Gaussian1D(1, 0, 0.1)
+    msg = "factor must have an integer value"
+    with pytest.raises(ValueError, match=msg):
+        discretize_model(gauss_1D, (-1, 2), mode="oversample", factor=1.2)

astropy/convolution/utils.py

@@ -120,15 +120,17 @@ def discretize_model(model, x_range, y_range=None, mode="center", factor=10):
                 bins. For 2D models, the interpolation is bilinear.
             * ``'oversample'``
                 Discretize model by taking the average of model values
-                on an oversampled grid.
+                in the pixel bins on an oversampled grid. Use the
+                ``factor`` keyword to set the integer oversampling
+                factor.
             * ``'integrate'``
                 Discretize model by integrating the model over the pixel
                 bins using `scipy.integrate.quad`. This mode conserves
                 the model integral on a subpixel scale, but is very
                 slow.
     factor : int, optional
-        The oversampling factor used when ``mode='oversample'``. Ignored
-        otherwise.
+        The integer oversampling factor used when ``mode='oversample'``.
+        Ignored otherwise.
 
     Returns
     -------
@@ -192,6 +194,10 @@ def discretize_model(model, x_range, y_range=None, mode="center", factor=10):
                 " 'y_range' must be a whole number."
             )
 
+    if factor != int(factor):
+        raise ValueError("factor must have an integer value")
+    factor = int(factor)
+
     if ndim == 2 and y_range is None:
         raise ValueError("y_range must be specified for a 2D model")
     if ndim == 1 and y_range is not None:

docs/changes/convolution/14794.api.rst

@@ -0,0 +1,2 @@
+``discretize_model`` will now raise a ``ValueError`` if
+``mode='oversample'`` and ``factor`` does not have an integer value.