BUGFIX: Argmax/argmin raywise needs to return int

spectral_cube/spectral_cube.py

@@ -795,7 +795,8 @@ def argmax(self, axis=None, how='auto', **kwargs):
         """
         return self.apply_numpy_function(np.nanargmax, fill=-np.inf,
                                          reduce=False, projection=False,
-                                         how=how, axis=axis, **kwargs)
+                                         how=how, axis=axis,
+                                         **kwargs)
 
     @aggregation_docstring
     @warn_slow
@@ -808,7 +809,8 @@ def argmin(self, axis=None, how='auto', **kwargs):
         """
         return self.apply_numpy_function(np.nanargmin, fill=np.inf,
                                          reduce=False, projection=False,
-                                         how=how, axis=axis, **kwargs)
+                                         how=how, axis=axis,
+                                         **kwargs)
 
     def _argmaxmin_world(self, axis, method, **kwargs):
         '''
@@ -997,7 +999,8 @@ def _cube_on_cube_operation(self, function, cube, equivalencies=[], **kwargs):
 
     def apply_function(self, function, axis=None, weights=None, unit=None,
                        projection=False, progressbar=False,
-                       update_function=None, keep_shape=False, **kwargs):
+                       update_function=None, keep_shape=False,
+                       **kwargs):
         """
         Apply a function to valid data along the specified axis or to the whole
         cube, optionally using a weight array that is the same shape (or at
@@ -1054,7 +1057,12 @@ def apply_function(self, function, axis=None, weights=None, unit=None,
         nz = self.shape[axis] if keep_shape else 1
 
         # allocate memory for output array
-        out = np.empty([nz, nx, ny]) * np.nan
+        # check dtype first (for argmax/argmin)
+        result = function(np.arange(3, dtype=self._data.dtype), **kwargs)
+        if 'int' in str(result.dtype):
+            out = np.zeros([nz, nx, ny], dtype=result.dtype)
+        else:
+            out = np.empty([nz, nx, ny]) * np.nan
 
         if progressbar:
             progressbar = ProgressBar(nx*ny)

spectral_cube/tests/test_spectral_cube.py

@@ -664,6 +664,16 @@ def test_argmin(self):
         self._check_numpy(self.c.argmin, d, np.nanargmin)
         self.c = self.d = None
 
+    def test_arg_rays(self, use_dask):
+        """
+        regression test: argmax must have integer dtype
+        """
+        if not use_dask:
+            result = self.c.argmax(how='ray')
+            assert 'int' in str(result.dtype)
+            result = self.c.argmin(how='ray')
+            assert 'int' in str(result.dtype)
+
     @pytest.mark.parametrize('iterate_rays', (True,False))
     def test_median(self, iterate_rays, use_dask):
         # Make sure that medians ignore empty/bad/NaN values