Make sure n_steps_default >= 3

ci/environment_docs.yml

@@ -15,3 +15,4 @@ dependencies:
   - myst-parser
   - myst-nb
   - sphinxcontrib-srclinks
+  - setuptools

gcm_filters/filter.py

@@ -71,6 +71,24 @@ def _taper_target(target_spec: TargetSpec):
 }
 
 
+def _compute_n_steps_default(
+    ndim, filter_shape, filter_scale, dx_min, transition_width
+):
+    """Compute the default number of steps for 1D or 2D filters based on provided parameters."""
+
+    n_steps_factor = filter_params[filter_shape][ndim]["offset"] + filter_params[
+        filter_shape
+    ][ndim]["factor"] * (
+        (np.pi / transition_width) ** filter_params[filter_shape][ndim]["exponent"]
+    )
+
+    filter_factor = filter_scale / dx_min
+
+    n_steps_default = max(np.ceil(n_steps_factor * filter_factor).astype(int), 3)
+
+    return n_steps_default
+
+
 class FilterSpec(NamedTuple):
     n_steps: int
     s_max: float
@@ -332,20 +350,19 @@ def __post_init__(self):
             raise ValueError(f"Transition width must be > 1.")
 
         # Get default number of steps
-        filter_factor = self.filter_scale / self.dx_min
         if self.ndim > 2:
             if self.n_steps < 3:
                 raise ValueError(f"When ndim > 2, you must set n_steps manually")
             else:
                 n_steps_default = self.n_steps  # For ndim>2 we don't have a default
         else:
-            n_steps_factor = filter_params[self.filter_shape][self.ndim][
-                "offset"
-            ] + filter_params[self.filter_shape][self.ndim]["factor"] * (
-                (np.pi / self.transition_width)
-                ** filter_params[self.filter_shape][self.ndim]["exponent"]
+            n_steps_default = _compute_n_steps_default(
+                self.ndim,
+                self.filter_shape,
+                self.filter_scale,
+                self.dx_min,
+                self.transition_width,
             )
-            n_steps_default = np.ceil(n_steps_factor * filter_factor).astype(int)
 
         # Set n_steps if needed and issue n_step warning, if needed
         if self.n_steps < 3:

readthedocs.yml

@@ -5,10 +5,5 @@ build:
   tools:
     python: "mambaforge-4.10"
 
-python:
-    install:
-       - method: setuptools
-         path: .
-
 conda:
   environment: ci/environment_docs.yml

tests/test_filter.py

@@ -7,7 +7,7 @@
 import xarray as xr
 
 from gcm_filters import Filter, FilterShape, GridType
-from gcm_filters.filter import FilterSpec
+from gcm_filters.filter import FilterSpec, _compute_n_steps_default
 
 
 def _check_equal_filter_spec(spec1, spec2):
@@ -85,6 +85,13 @@ def test_filter_spec(filter_args, expected_filter_spec):
     # TODO: check other properties of filter_spec?
 
 
+def test_default_n_steps_larger_equal_3():
+    n_steps_default = _compute_n_steps_default(2, FilterShape.GAUSSIAN, 1.5, 1, np.pi)
+
+    # Assert that the number of steps is greater than or equal to 3
+    assert n_steps_default >= 3
+
+
 #################### Diffusion-based filter tests ########################################
 area_weighted_regular_grids = [
     GridType.REGULAR_AREA_WEIGHTED,