Eoepca test suite (#297)

* array concat handle axis

* array append handle axis

* fix array apply cases

* update for tests

* array test update

* fix array apply test

Author: ValentinaHutter

openeo_processes_dask/process_implementations/arrays.py

@@ -50,7 +50,7 @@
 ]
 
 
-def get_labels(data, dimension="labels", axis=0):
+def get_labels(data, dimension="labels", axis=0, dim_labels=None):
     if isinstance(data, xr.DataArray):
         dimension = data.dims[0] if len(data.dims) == 1 else dimension
         if axis:
@@ -61,6 +61,8 @@ def get_labels(data, dimension="labels", axis=0):
         labels = []
         if isinstance(data, list):
             data = np.asarray(data)
+    if not isinstance(dim_labels, type(None)):
+        labels = dim_labels
     return labels, data
 
 
@@ -82,9 +84,7 @@ def array_element(
         raise ArrayElementParameterConflict(
             "The process `array_element` only allows that either the `index` or the `labels` parameter is set."
         )
-
-    if isinstance(data, xr.DataArray):
-        dim_labels, data = get_labels(data, axis=axis)
+    dim_labels, data = get_labels(data, axis=axis, dim_labels=dim_labels)
 
     if label is not None:
         if len(dim_labels) == 0:
@@ -189,7 +189,7 @@ def modify(data):
     return modified
 
 
-def array_concat(array1: ArrayLike, array2: ArrayLike) -> ArrayLike:
+def array_concat(array1: ArrayLike, array2: ArrayLike, axis=None) -> ArrayLike:
     labels1, array1 = get_labels(array1)
     labels2, array2 = get_labels(array2)
 
@@ -198,7 +198,21 @@ def array_concat(array1: ArrayLike, array2: ArrayLike) -> ArrayLike:
             "At least one label exists in both arrays and the conflict must be resolved before."
         )
 
-    concat = np.concatenate([array1, array2])
+    if (len(array1.shape) - len(array2.shape)) == 1:
+        if axis is None:
+            s1 = np.array(list(array1.shape))
+            s2 = list(array2.shape)
+            s2.append(0)
+            s2 = np.array(s2)
+
+            axis = np.argmax(s1 != s2)
+
+        array2 = np.expand_dims(array2, axis=axis)
+
+    if axis:
+        concat = np.concatenate([array1, array2], axis=axis)
+    else:
+        concat = np.concatenate([array1, array2])
 
     # e.g. concating int32 and str arrays results in the result being cast to a Unicode dtype of a certain length (e.g. <U22).
     # There isn't really anything better to do as numpy does not support heterogenuous arrays.
@@ -219,7 +233,20 @@ def array_append(
     value: Any,
     label: Optional[Any] = None,
     dim_labels=None,
+    axis=None,
 ) -> ArrayLike:
+    if axis:
+        if isinstance(value, list) and len(value) == 1:
+            value = value[0]
+        if (isinstance(value, np.ndarray) or isinstance(value, da.core.Array)) and len(
+            value.flatten()
+        ) == 1:
+            value = value.flatten()[0]
+
+        value = np.take(np.ones_like(data), indices=0, axis=axis) * value
+        concat = array_concat(data, value, axis=axis)
+        return concat
+
     if dim_labels:
         data = array_create_labeled(data=data, labels=dim_labels)
     if label is not None:

openeo_processes_dask/process_implementations/cubes/apply.py

@@ -77,22 +77,45 @@ def apply_dimension(
         exclude_dims={dimension},
     )
 
-    reordered_result = result.transpose(*data.dims, ...).rename(
-        {dimension: target_dimension}
-    )
+    reordered_result = result.transpose(*data.dims, ...)
 
-    if len(reordered_result[target_dimension]) == 1:
-        reordered_result[target_dimension] = ["0"]
+    if dimension in reordered_result.dims:
+        result_len = len(reordered_result[dimension])
+    else:
+        result_len = 1
+
+    # Case 1: target_dimension is not defined/ is source dimension
+    if dimension == target_dimension:
+        # dimension labels preserved
+        # if the number of source dimension's values is equal to the number of computed values
+        if len(reordered_data[dimension]) == result_len:
+            reordered_result[dimension] == reordered_data[dimension].values
+        else:
+            reordered_result[dimension] = np.arange(result_len)
+    elif target_dimension in reordered_result.dims:
+        # source dimension is not target dimension
+        # target dimension exists with a single label only
+        if len(reordered_result[target_dimension]) == 1:
+            reordered_result = reordered_result.drop_vars(target_dimension).squeeze(
+                target_dimension
+            )
+            reordered_result = reordered_result.rename({dimension: target_dimension})
+            reordered_result[dimension] = np.arange(result_len)
+        else:
+            raise Exception(
+                f"Cannot rename dimension {dimension} to {target_dimension} as {target_dimension} already exists in dataset and contains more than one label: {reordered_result[target_dimension]}. See process definition. "
+            )
+    else:
+        # source dimension is not the target dimension and the latter does not exist
+        reordered_result = reordered_result.rename({dimension: target_dimension})
+        reordered_result[target_dimension] = np.arange(result_len)
 
     if data.rio.crs is not None:
         try:
             reordered_result.rio.write_crs(data.rio.crs, inplace=True)
         except ValueError:
             pass
 
-    if is_new_dim_added:
-        reordered_result.openeo.add_dim_type(name=target_dimension, type="other")
-
     return reordered_result
 
 

tests/test_apply.py

@@ -47,7 +47,7 @@ def test_apply(temporal_interval, bounding_box, random_raster_data, process_regi
 
 @pytest.mark.parametrize("size", [(6, 5, 4, 4)])
 @pytest.mark.parametrize("dtype", [np.float32])
-def test_apply_dimension_case_1(
+def test_apply_dimension_add(
     temporal_interval, bounding_box, random_raster_data, process_registry
 ):
     input_cube = create_fake_rastercube(
@@ -78,61 +78,6 @@ def test_apply_dimension_case_1(
     )
 
 
-@pytest.mark.parametrize("size", [(6, 5, 4, 4)])
-@pytest.mark.parametrize("dtype", [np.float32])
-def test_apply_dimension_target_dimension(
-    temporal_interval, bounding_box, random_raster_data, process_registry
-):
-    input_cube = create_fake_rastercube(
-        data=random_raster_data,
-        spatial_extent=bounding_box,
-        temporal_extent=temporal_interval,
-        bands=["B02", "B03", "B04", "B08"],
-        backend="dask",
-    )
-
-    _process = partial(
-        process_registry["mean"].implementation,
-        data=ParameterReference(from_parameter="data"),
-    )
-
-    # Target dimension is null and therefore defaults to the source dimension
-    output_cube_reduced = apply_dimension(
-        data=input_cube, process=_process, dimension="x", target_dimension="target"
-    )
-
-    expected_output = (input_cube.mean(dim="x")).expand_dims("target")
-
-    general_output_checks(
-        input_cube=input_cube,
-        output_cube=output_cube_reduced,
-        verify_attrs=True,
-        verify_crs=False,
-        expected_results=expected_output,
-    )
-
-    # Target dimension is null and therefore defaults to the source dimension
-    output_cube_reduced = apply_dimension(
-        data=input_cube, process=_process, dimension="x", target_dimension="y"
-    )
-    expected_output = (
-        input_cube.mean(dim="x")
-        .expand_dims("target")
-        .drop_vars("y")
-        .rename({"target": "y"})
-    )
-
-    general_output_checks(
-        input_cube=input_cube,
-        output_cube=output_cube_reduced,
-        verify_attrs=True,
-        verify_crs=False,
-        expected_results=expected_output,
-    )
-
-    assert "y" in output_cube_reduced.openeo.other_dims
-
-
 @pytest.mark.parametrize("size", [(6, 5, 4, 4)])
 @pytest.mark.parametrize("dtype", [np.float32])
 def test_apply_dimension_ordering_processes(