Additional options for sign_agreement (#118)

* More file handling functions.

* Added option to not count nans in sign_agreement

* Fixed default value

* Update whats-new.rst

docs/whats-new.rst

@@ -2,6 +2,13 @@
 
 What's New
 ===========
+.. _whats-new.1.1.0:
+
+v1.1.0 (unreleased)
+
+New Features
+~~~~~~~~~~~~
+- `sing_agreement` now supports the option to not count nans along the given dimension (:pull:`118`). By `Julius Busecke <https://github.com/jbusecke>`_
 
 .. _whats-new.1.0.0:
 

xarrayutils/test/test_utils.py

@@ -273,6 +273,22 @@ def test_linear_trend():
  assert np.allclose(fit, test)
 
 
+def test_sign_agreement_count_nans():
+ a = xr.DataArray(
+ [
+ [1, -1, np.nan, 4],
+ [1, 1, -1, 4],
+ [1, -1, -1, 4],
+ [1, 1, -1, 4],
+ [1, -1, -1, 4],
+ ],
+ dims=["i", "j"],
+ )
+ sa = sign_agreement(a, a.mean("i"), "i", count_nans=False)
+ expected = xr.DataArray([1.0, 0.0, 1.0, 1.0], dims=["j"])
+ xr.testing.assert_allclose(expected, sa)
+
+
 def test_sign_agreement():
  # test dataset
  target_dim = "member"

xarrayutils/utils.py

@@ -729,7 +729,7 @@ def lag_and_combine(ds, lags, dim="time"):
  return xr.concat(datasets, dim=concat_dim_da(lags, "lag"))
 
 
-def sign_agreement(da, ds_ref, dim, threshold=0.75, mask=True):
+def sign_agreement(da, ds_ref, dim, threshold=0.75, mask=True, count_nans=True):
  """[summary]
 
  Parameters
@@ -743,12 +743,18 @@ def sign_agreement(da, ds_ref, dim, threshold=0.75, mask=True):
  threshold : float, optional
  The minimum fraction of elements that have to agree along `dim`, by default 0.75 (75%)
  mask : bool, optional
- If True nan values get masked out in the output, by default True
+ If True, datapoints with all nan values along `dim` get masked out in the output, by default True
+ count_nans : bool, optional
+ If True, nans along `dim` are counted towards the threshold. If False sign agreement is
+ calculated according to non-nan values only, by default True
 
  """
  if mask:
  mask_data = np.isnan(da).all(dim)
- ndim = len(da[dim].data)
+ if count_nans:
+ ndim = len(da[dim].data)
+ else:
+ ndim = (~np.isnan(da)).sum(dim)
  sign_agreement = (np.sign(da) == np.sign(ds_ref)).sum(dim) >= (threshold * ndim)
  if mask:
  sign_agreement = sign_agreement.where(~mask_data)