Merge pull request #67 from efiring/array_ufunc

Handle xarray DataArray in wrapped ufuncs

Author: efiring

.travis.yml

@@ -36,7 +36,7 @@ before_install:
     conda create --name TEST python=$PY --file requirements-dev.txt --quiet
     source activate TEST
     # Install after to ensure it will be downgraded when testing an older version.
-    conda install numpy=$NUMPY
+    conda install numpy=$NUMPY xarray dask
     conda info --all
 
 # Test source distribution.
@@ -64,7 +64,7 @@ script:
       pushd docs
       make clean html linkcheck
       popd
-      if [[ -z "$TRAVIS_TAG" ]]; then 
+      if [[ -z "$TRAVIS_TAG" ]]; then
         python -m doctr deploy --build-tags --key-path github_deploy_key.enc --built-docs docs/_build/html dev
       else
         python -m doctr deploy --build-tags --key-path github_deploy_key.enc --built-docs docs/_build/html "version-$TRAVIS_TAG"

gsw/_utilities.py

@@ -26,20 +26,35 @@ def wrapper(*args, **kw):
             args = list(args)
             args.append(p)
 
-        isarray = np.any([hasattr(a, '__iter__') for a in args])
-        ismasked = np.any([np.ma.isMaskedArray(a) for a in args])
+        isarray = [hasattr(a, '__iter__') for a in args]
+        ismasked = [np.ma.isMaskedArray(a) for a in args]
+        isduck = [hasattr(a, '__array_ufunc__')
+                    and not isinstance(a, np.ndarray) for a in args]
+
+        hasarray = np.any(isarray)
+        hasmasked = np.any(ismasked)
+        hasduck = np.any(isduck)
 
         def fixup(ret):
-            if ismasked:
+            if hasduck:
+                return ret
+            if hasmasked:
                 ret = np.ma.masked_invalid(ret)
-            if not isarray and isinstance(ret, np.ndarray):
-                ret = ret[0]
+            if not hasarray and isinstance(ret, np.ndarray) and ret.size == 1:
+                try:
+                    ret = ret[0]
+                except IndexError:
+                    pass
             return ret
 
-        if ismasked:
-            newargs = [masked_to_nan(a) for a in args]
-        else:
-            newargs = [np.asarray(a, dtype=float) for a in args]
+        newargs = []
+        for i, arg in enumerate(args):
+            if ismasked[i]:
+                newargs.append(masked_to_nan(arg))
+            elif isduck[i]:
+                newargs.append(arg)
+            else:
+                newargs.append(np.asarray(arg, dtype=float))
 
         if p is not None:
             kw['p'] = newargs.pop()

gsw/geostrophy.py

@@ -73,7 +73,10 @@ def geo_strf_dyn_height(SA, CT, p, p_ref=0, axis=0, max_dp=1.0,
     dh = np.empty(SA.shape, dtype=float)
     dh.fill(np.nan)
 
-    order = 'F' if SA.flags.fortran else 'C'
+    try:
+        order = 'F' if SA.flags.fortran else 'C'
+    except AttributeError:
+        order = 'C'  # e.g., xarray DataArray doesn't have flags
     for ind in indexer(SA.shape, axis, order=order):
         igood = goodmask[ind]
         # If p_ref is below the deepest value, skip the profile.

gsw/tests/check_functions.py

@@ -55,7 +55,7 @@ def find(x):
     """
     Numpy equivalent to Matlab find.
     """
-    return np.nonzero(x.flatten())[0]
+    return np.nonzero(np.asarray(x).flatten())[0]
 
 
 def group_or(line):

gsw/tests/test_xarray.py

@@ -0,0 +1,143 @@
+"""
+Tests functions with xarray inputs.
+
+This version is a copy of the original test_check_functions but with
+an import of xarray, and conversion of the 3 main check cast arrays
+into DataArray objects.
+
+An additional xarray-dask test is added.
+"""
+
+import os
+import pytest
+
+import numpy as np
+from numpy.testing import assert_allclose
+
+import gsw
+from gsw._utilities import Bunch
+from check_functions import parse_check_functions
+
+xr = pytest.importorskip('xarray')
+
+# Most of the tests have some nan values, so we need to suppress the warning.
+# Any more careful fix would likely require considerable effort.
+np.seterr(invalid='ignore')
+
+root_path = os.path.abspath(os.path.dirname(__file__))
+
+# Function checks that we can't handle automatically yet.
+blacklist = ['deltaSA_atlas',  # the test is complicated; doesn't fit the pattern.
+             'geostrophic_velocity',  # test elsewhere; we changed the API
+             #'CT_from_entropy', # needs prior entropy_from_CT; don't have it in C
+             #'CT_first_derivatives', # passes, but has trouble in "details";
+                                      # see check_functions.py
+             #'entropy_second_derivatives', # OK now; handling extra parens.
+             #'melting_ice_into_seawater',  # OK now; fixed nargs mismatch.
+             ]
+
+# We get an overflow from ct_from_enthalpy_exact, but the test passes.
+cv = Bunch(np.load(os.path.join(root_path, 'gsw_cv_v3_0.npz')))
+
+# Substitute new check values for the pchip interpolation version.
+cv.geo_strf_dyn_height = np.load(os.path.join(root_path,'geo_strf_dyn_height.npy'))
+cv.geo_strf_velocity = np.load(os.path.join(root_path,'geo_strf_velocity.npy'))
+
+for name in ['SA_chck_cast', 't_chck_cast', 'p_chck_cast']:
+    cv[name] = xr.DataArray(cv[name])
+
+cf = Bunch()
+
+d = dir(gsw)
+funcnames = [name for name in d if '__' not in name]
+
+mfuncs = parse_check_functions(os.path.join(root_path, 'gsw_check_functions_save.m'))
+mfuncs = [mf for mf in mfuncs if mf.name in d and mf.name not in blacklist]
+mfuncnames = [mf.name for mf in mfuncs]
+
+
+@pytest.fixture(scope='session', params=mfuncs)
+def cfcf(request):
+    return cv, cf, request.param
+
+
+def test_check_function(cfcf):
+    cv, cf, mfunc = cfcf
+    mfunc.run(locals())
+    if mfunc.exception is not None or not mfunc.passed:
+        print('\n', mfunc.name)
+        print('  ', mfunc.runline)
+        print('  ', mfunc.testline)
+        if mfunc.exception is None:
+            mfunc.exception = ValueError('Calculated values are different from the expected matlab results.')
+        raise mfunc.exception
+    else:
+        print(mfunc.name)
+        assert mfunc.passed
+
+
+def test_dask_chunking():
+    dsa = pytest.importorskip('dask.array')
+
+    # define some input data
+    shape = (100, 1000)
+    chunks = (100, 200)
+    sp = xr.DataArray(dsa.full(shape, 35., chunks=chunks), dims=['time', 'depth'])
+    p = xr.DataArray(np.arange(shape[1]), dims=['depth'])
+    lon = 0
+    lat = 45
+
+    sa = gsw.SA_from_SP(sp, p, lon, lat)
+    sa_dask = sa.compute()
+
+    sa_numpy = gsw.SA_from_SP(np.full(shape, 35.0), p.values, lon, lat)
+    assert_allclose(sa_dask, sa_numpy)
+
+
+# Additional tests from Graeme MacGilchrist
+# https://nbviewer.jupyter.org/github/gmacgilchrist/wmt_bgc/blob/master/notebooks/test_gsw-xarray.ipynb
+
+# Define dimensions and coordinates
+dims = ['y','z','t']
+# 2x2x2
+y = np.arange(0,2)
+z = np.arange(0,2)
+t = np.arange(0,2)
+# Define numpy arrays of salinity, temperature and pressure
+SA_vals = np.array([[[34.7,34.8],[34.9,35]],[[35.1,35.2],[35.3,35.4]]])
+CT_vals = np.array([[[7,8],[9,10]],[[11,12],[13,14]]])
+p_vals = np.array([10,20])
+lat_vals = np.array([0,10])
+# Plug in to xarray objects
+SA = xr.DataArray(SA_vals,dims=dims,coords={'y':y,'z':z,'t':t})
+CT = xr.DataArray(CT_vals,dims=dims,coords={'y':y,'z':z,'t':t})
+p = xr.DataArray(p_vals,dims=['z'],coords={'z':z})
+lat = xr.DataArray(lat_vals,dims=['y'],coords={'y':y})
+
+
+def test_xarray_with_coords():
+    pytest.importorskip('dask')
+    SA_chunk = SA.chunk(chunks={'y':1,'t':1})
+    CT_chunk = CT.chunk(chunks={'y':1,'t':1})
+    lat_chunk = lat.chunk(chunks={'y':1})
+
+    # Dimensions and cordinates match:
+    expected = gsw.sigma0(SA_vals, CT_vals)
+    xarray = gsw.sigma0(SA, CT)
+    chunked = gsw.sigma0(SA_chunk, CT_chunk)
+    assert_allclose(xarray, expected)
+    assert_allclose(chunked, expected)
+
+    # Broadcasting along dimension required (dimensions known)
+    expected = gsw.alpha(SA_vals, CT_vals, p_vals[np.newaxis, :, np.newaxis])
+    xarray = gsw.alpha(SA, CT, p)
+    chunked = gsw.alpha(SA_chunk, CT_chunk, p)
+    assert_allclose(xarray, expected)
+    assert_allclose(chunked, expected)
+
+    # Broadcasting along dimension required (dimensions unknown/exclusive)
+    expected = gsw.z_from_p(p_vals[:, np.newaxis], lat_vals[np.newaxis, :])
+    xarray = gsw.z_from_p(p, lat)
+    chunked = gsw.z_from_p(p,lat_chunk)
+    assert_allclose(xarray, expected)
+    assert_allclose(chunked, expected)