update array processes (#280)

* update array_find

* update array contains

* add array_interpolate

* add array_interpolate

* add array_interpolate

* update tests!

openeo_processes_dask/process_implementations/arrays.py

@@ -33,6 +33,7 @@
     "array_find",
     "array_labels",
     "array_apply",
+    "array_interpolate_linear",
     "first",
     "last",
     "order",
@@ -165,10 +166,8 @@ def array_contains(data: ArrayLike, value: Any, axis=None) -> bool:
             value_is_valid = True
     if len(np.shape(data)) != 1 and axis is None:
         return False
-    if not value_is_valid:
+    if not value_is_valid or pd.isnull(value):
         return False
-    if pd.isnull(value):
-        return np.isnan(data).any(axis=axis)
     else:
         return np.isin(data, value).any(axis=axis)
 
@@ -188,8 +187,14 @@ def array_find(
     idxs = (data == value).argmax(axis=axis)
 
     mask = ~np.array((data == value).any(axis=axis))
-    if np.isnan(value):
+    if not isinstance(value, str) and np.isnan(value):
         mask = True
+    if reverse:
+        if axis is None:
+            size = data.size
+        else:
+            size = data.shape[axis]
+        idxs = size - 1 - idxs
 
     logger.warning(
         "array_find: numpy has no sentinel value for missing data in integer arrays, therefore np.masked_array is used to return the indices of found elements. Further operations might fail if not defined for masked arrays."
@@ -230,6 +235,19 @@ def array_apply(
         )
 
 
+def array_interpolate_linear(data: ArrayLike):
+    if isinstance(data, list):
+        data = np.array(data)
+    x = np.arange(len(data))
+    valid = np.isfinite(data)
+    if len(x[valid]) < 2:
+        return data
+    data[~valid] = np.interp(
+        x[~valid], x[valid], data[valid], left=np.nan, right=np.nan
+    )
+    return data
+
+
 def first(
     data: ArrayLike,
     ignore_nodata: Optional[bool] = True,

tests/test_arrays.py

@@ -180,7 +180,7 @@ def test_array_append(data, value, expected):
         ([1, 2, 3], 2, True),
         (["A", "B", "C"], "b", False),
         ([1, 2, 3], "2", False),
-        ([1, 2, np.nan], np.nan, True),
+        ([1, 2, np.nan], np.nan, False),
         ([[2, 1], [3, 4]], [1, 2], False),
         ([[2, 1], [3, 4]], 2, False),
         ([1, 2, 3], np.int64(2), True),
@@ -222,14 +222,15 @@ def test_array_contains_object_dtype():
     [
         ([1, 0, 3, 2], 3, 2, None, False),
         ([1, 0, 3, 2, np.nan, 3], np.nan, 999999, None, False),
-        ([1, 0, 3, 2], 3, 2, None, False),
+        ([1, 0, 3, 0, 2], 0, 1, None, False),
         ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [999999, 1, 0, 999999], 0, False),
         ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [2, 1], 1, False),
-        ([1, 0, 3, 2], 3, 1, None, True),
+        ([1, 0, 3, 2], 3, 2, None, True),
         ([1, 0, 3, 2, np.nan, 3], np.nan, 999999, None, True),
-        ([1, 0, 3, 2], 3, 1, None, True),
-        ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [999999, 0, 1, 999999], 0, True),
-        ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [1, 2], 1, True),
+        ([1, 0, 3, 0, 2], 0, 3, None, True),
+        ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [999999, 1, 0, 999999], 0, True),
+        ([[1, 0, 3, 2], [5, 3, 6, 8]], 3, [2, 1], 1, True),
+        (["A", "B", "C"], "b", 99999, None, False),
     ],
 )
 def test_array_find(data, value, expected, axis, reverse):
@@ -263,6 +264,33 @@ def test_array_apply(process_registry):
     assert (output_cube == [2, 3, 4, 5, 6, 7]).all()
 
 
+@pytest.mark.parametrize(
+    "data, expected",
+    [
+        ([np.nan, 1, np.nan, 6, np.nan, -8], [np.nan, 1, 3.5, 6, -1, -8]),
+        ([np.nan, 1, np.nan, np.nan], [np.nan, 1, np.nan, np.nan]),
+    ],
+)
+def test_array_interpolate_linear(data, expected):
+    assert np.array_equal(
+        array_interpolate_linear(data),
+        expected,
+        equal_nan=True,
+    )
+    data_np = np.array(data)
+    assert np.array_equal(
+        array_interpolate_linear(data_np),
+        expected,
+        equal_nan=True,
+    )
+    data_da = da.from_array(data_np)
+    assert np.array_equal(
+        array_interpolate_linear(data_da),
+        expected,
+        equal_nan=True,
+    )
+
+
 def test_first():
     assert first(np.array([1, 0, 3, 2])) == 1
     assert pd.isnull(first(np.array([np.nan, 2, 3]), ignore_nodata=False))