Let TensorFlow accessor and data loader handle either xarray.DataArray or xarray.Dataset inputs

xbatcher/accessors.py

@@ -1,8 +1,23 @@
+from typing import Union
+
 import xarray as xr
 
 from .generators import BatchGenerator
 
 
+def _as_xarray_dataarray(
+    xr_obj: Union[xr.Dataset, xr.DataArray]
+) -> Union[xr.Dataset, xr.DataArray]:
+    """
+    Convert xarray.Dataset to xarray.DataArray if needed, so that it can
+    be converted into a torch.Tensor object.
+    """
+    if isinstance(xr_obj, xr.Dataset):
+        xr_obj = xr_obj.to_array().squeeze(dim="variable")
+
+    return xr_obj
+
+
 @xr.register_dataarray_accessor("batch")
 @xr.register_dataset_accessor("batch")
 class BatchAccessor:
@@ -26,31 +41,32 @@ def generator(self, *args, **kwargs):
         return BatchGenerator(self._obj, *args, **kwargs)
 
 
+@xr.register_dataarray_accessor("keras")
+@xr.register_dataset_accessor("keras")
+class KerasAccessor:
+    def __init__(self, xarray_obj):
+        self._obj = xarray_obj
+
+    def to_tensor(self):
+        """Convert this DataArray to a tensorflow.Tensor"""
+        import tensorflow as tf
+
+        dataarray = _as_xarray_dataarray(xr_obj=self._obj)
+
+        return tf.convert_to_tensor(dataarray.data)
+
+
 @xr.register_dataarray_accessor("torch")
 @xr.register_dataset_accessor("torch")
 class TorchAccessor:
     def __init__(self, xarray_obj):
         self._obj = xarray_obj
 
-    def _as_xarray_dataarray(self, xr_obj):
-        """
-        Convert xarray.Dataset to xarray.DataArray if needed, so that it can
-        be converted into a torch.Tensor object.
-        """
-        try:
-            # Convert xr.Dataset to xr.DataArray
-            dataarray = xr_obj.to_array().squeeze(dim="variable")
-        except AttributeError:  # 'DataArray' object has no attribute 'to_array'
-            # If object is already an xr.DataArray
-            dataarray = xr_obj
-
-        return dataarray
-
     def to_tensor(self):
         """Convert this DataArray to a torch.Tensor"""
         import torch
 
-        dataarray = self._as_xarray_dataarray(xr_obj=self._obj)
+        dataarray = _as_xarray_dataarray(xr_obj=self._obj)
 
         return torch.tensor(data=dataarray.data)
 
@@ -62,6 +78,6 @@ def to_named_tensor(self):
         """
         import torch
 
-        dataarray = self._as_xarray_dataarray(xr_obj=self._obj)
+        dataarray = _as_xarray_dataarray(xr_obj=self._obj)
 
         return torch.tensor(data=dataarray.data, names=tuple(dataarray.sizes))

xbatcher/tests/test_accessors.py

@@ -22,6 +22,30 @@ def sample_ds_3d():
     return ds
 
 
+@pytest.fixture(scope="module")
+def sample_dataArray():
+    return xr.DataArray(np.zeros((2, 4), dtype="i4"), dims=("x", "y"), name="foo")
+
+
+@pytest.fixture(scope="module")
+def sample_Dataset():
+    return xr.Dataset(
+        {
+            "x": xr.DataArray(np.arange(10), dims="x"),
+            "foo": xr.DataArray(np.ones(10, dtype="float"), dims="x"),
+        }
+    )
+
+
+def test_as_xarray_dataarray(sample_dataArray, sample_Dataset):
+    assert isinstance(
+        xbatcher.accessors._as_xarray_dataarray(sample_dataArray), xr.DataArray
+    )
+    assert isinstance(
+        xbatcher.accessors._as_xarray_dataarray(sample_Dataset), xr.DataArray
+    )
+
+
 def test_batch_accessor_ds(sample_ds_3d):
     bg_class = BatchGenerator(sample_ds_3d, input_dims={"x": 5})
     bg_acc = sample_ds_3d.batch.generator(input_dims={"x": 5})
@@ -40,6 +64,25 @@ def test_batch_accessor_da(sample_ds_3d):
         assert batch_class.equals(batch_acc)
 
 
+@pytest.mark.parametrize(
+    "foo_var",
+    [
+        "foo",  # xr.DataArray
+        ["foo"],  # xr.Dataset
+    ],
+)
+def test_keras_to_tensor(sample_ds_3d, foo_var):
+    tensorflow = pytest.importorskip("tensorflow")
+
+    foo = sample_ds_3d[foo_var]
+    t = foo.keras.to_tensor()
+    assert isinstance(t, tensorflow.Tensor)
+    assert t.shape == tuple(foo.sizes.values())
+
+    foo_array = foo.to_array().squeeze() if hasattr(foo, "to_array") else foo
+    np.testing.assert_array_equal(t, foo_array.values)
+
+
 @pytest.mark.parametrize(
     "foo_var",
     [