test(pca_test): relax nan and nodata tests

The results sometimes deviate in order (#454). The problematic tests
were relaxed to expect this. Also refactored other test code to decrease
code repetition.

Author: nialov

tests/exploratory_analyses/pca_test.py

@@ -1,163 +1,185 @@
 import sys
-from pathlib import Path
 
 import geopandas as gpd
 import numpy as np
 import pandas as pd
 import pytest
+from beartype.typing import Optional
 from shapely.geometry import Point
 
 from eis_toolkit.exceptions import EmptyDataException, InvalidColumnException, InvalidParameterValueException
 from eis_toolkit.exploratory_analyses.pca import compute_pca
 
-parent_dir = Path(__file__).parent
-MULTIBAND_RASTER_PATH = parent_dir.joinpath("../data/remote/small_raster_multiband.tif")
-
 DATA = np.array([[1, 1], [2, 2], [3, 3]])
+EXPECTED_DATA_PCA_VALUES = expected_pca_values = np.array(
+    [[-1.73205081, 1.11022302e-16], [0.0, 0.0], [1.73205081, 1.11022302e-16]]
+)
+EXPECTED_DATA_COMPONENT_VALUES = np.array([[0.70711, 0.70711], [0.70711, -0.70711]])
+EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE = np.array([[0.70711, 0.70711], [-0.70711, 0.70711]])
+EXPECTED_DATA_EXPLAINED_VARIANCE_RATIOS_VALUES = [1.0, 4.10865055e-33]
+
+DATA_DF = pd.DataFrame(data=DATA, columns=["A", "B"])
+EXPECTED_DATA_DF_COLUMNS = ["principal_component_1", "principal_component_2"]
+
+DATA_GDF = gpd.GeoDataFrame(
+    data=DATA, columns=["A", "B"], geometry=[Point(1, 2), Point(2, 1), Point(3, 3)], crs="EPSG:4326"
+)
+EXPECTED_DATA_GDF_COLUMNS = ["principal_component_1", "principal_component_2", "geometry"]
+
+DATA_WITH_NAN = np.array([[1, 1], [2, np.nan], [3, 3]])
+DATA_WITH_NODATA = np.array([[1, 1], [2, -9999], [3, 3]])
+
+
+def _assert_expected_values(
+    pca_array: np.ndarray,
+    principal_components,
+    explained_variances,
+    explained_variance_ratios,
+    expected_pca_values=EXPECTED_DATA_PCA_VALUES,
+    expected_component_values=EXPECTED_DATA_COMPONENT_VALUES,
+    expected_component_values_alternative: Optional[np.ndarray] = None,
+    expected_explained_variance_ratios_values=EXPECTED_DATA_EXPLAINED_VARIANCE_RATIOS_VALUES,
+    decimal_accuracy: int = 5,
+    data_shape=DATA.shape,
+):
+    np.testing.assert_equal(principal_components.size, 4)
+    np.testing.assert_equal(explained_variances.size, 2)
+    np.testing.assert_equal(explained_variance_ratios.size, 2)
+    np.testing.assert_equal(pca_array.shape, data_shape)
+    np.testing.assert_array_almost_equal(pca_array, expected_pca_values, decimal=decimal_accuracy)
+
+    try:
+        np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=decimal_accuracy)
+    except AssertionError:
+        # Both variations in the sign of the two last members of principal_components occurs
+        # depending on environment in nan and nodata tests
+        # Both are allowed for those
+        if expected_component_values_alternative is None:
+            # Deviations in order are not expected unless *_alternative array is passed as input
+            raise
+        np.testing.assert_array_almost_equal(
+            principal_components, expected_component_values_alternative, decimal=decimal_accuracy
+        )
+
+    np.testing.assert_array_almost_equal(
+        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=decimal_accuracy
+    )
 
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_numpy_array():
     """Test that PCA function gives correct output for Numpy array input."""
     pca_array, principal_components, explained_variances, explained_variance_ratios = compute_pca(DATA, 2)
 
-    expected_pca_array_values = np.array([[-1.73205081, 1.11022302e-16], [0.0, 0.0], [1.73205081, 1.11022302e-16]])
-    expected_component_values = np.array([[0.70711, 0.70711], [0.70711, -0.70711]])
-    expected_explained_variance_ratios_values = [1.0, 4.10865055e-33]
-
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(pca_array.shape, DATA.shape)
-
-    np.testing.assert_array_almost_equal(pca_array, expected_pca_array_values, decimal=5)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=5)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=5
+    _assert_expected_values(
+        pca_array=pca_array,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
     )
 
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_df():
     """Test that PCA function gives correct output for DF input."""
-    data_df = pd.DataFrame(data=DATA, columns=["A", "B"])
 
-    pca_df, principal_components, explained_variances, explained_variance_ratios = compute_pca(data_df, 2)
+    pca_df, principal_components, explained_variances, explained_variance_ratios = compute_pca(DATA_DF, 2)
 
-    expected_columns = ["principal_component_1", "principal_component_2"]
-    expected_pca_values = np.array([[-1.73205081, 1.11022302e-16], [0.0, 0.0], [1.73205081, 1.11022302e-16]])
-    expected_component_values = np.array([[0.70711, 0.70711], [0.70711, -0.70711]])
-    expected_explained_variance_ratios_values = [1.0, 4.10865055e-33]
-
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(list(pca_df.columns), expected_columns)
-    np.testing.assert_equal(pca_df.shape, data_df.shape)
-
-    np.testing.assert_array_almost_equal(pca_df.values, expected_pca_values, decimal=5)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=5)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=5
+    _assert_expected_values(
+        pca_array=pca_df.values,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
     )
+    np.testing.assert_equal(list(pca_df.columns), EXPECTED_DATA_DF_COLUMNS)
+    np.testing.assert_equal(pca_df.shape, DATA_DF.shape)
 
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_gdf():
     """Test that PCA function gives correct output for GDF input."""
-    data_gdf = gpd.GeoDataFrame(
-        data=DATA, columns=["A", "B"], geometry=[Point(1, 2), Point(2, 1), Point(3, 3)], crs="EPSG:4326"
-    )
-
-    pca_gdf, principal_components, explained_variances, explained_variance_ratios = compute_pca(data_gdf, 2)
 
-    expected_columns = ["principal_component_1", "principal_component_2", "geometry"]
-    expected_pca_values = np.array([[-1.73205081, 1.11022302e-16], [0.0, 0.0], [1.73205081, 1.11022302e-16]])
-    expected_component_values = np.array([[0.70711, 0.70711], [0.70711, -0.70711]])
-    expected_explained_variance_ratios_values = [1.0, 4.10865055e-33]
+    pca_gdf, principal_components, explained_variances, explained_variance_ratios = compute_pca(DATA_GDF, 2)
 
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(list(pca_gdf.columns), expected_columns)
-    np.testing.assert_equal(pca_gdf.shape, data_gdf.shape)
-
-    np.testing.assert_array_almost_equal(pca_gdf.drop(columns=["geometry"]).values, expected_pca_values, decimal=5)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=5)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=5
+    _assert_expected_values(
+        pca_array=pca_gdf.drop(columns=["geometry"]).values,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
     )
 
+    np.testing.assert_equal(list(pca_gdf.columns), EXPECTED_DATA_GDF_COLUMNS)
+    np.testing.assert_equal(pca_gdf.shape, DATA_GDF.shape)
+
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_with_nan_removal():
     """Test that PCA function gives correct output for Numpy array input that has NaN values and remove strategy."""
-    data = np.array([[1, 1], [2, np.nan], [3, 3]])
     pca_array, principal_components, explained_variances, explained_variance_ratios = compute_pca(
-        data, 2, nodata_handling="remove"
+        DATA_WITH_NAN, 2, nodata_handling="remove"
     )
 
     expected_pca_values = np.array([[-1.414, 0.0], [np.nan, np.nan], [1.414, 0.0]])
-    expected_component_values = np.array([[0.70711, 0.70711], [-0.70711, 0.70711]])
     expected_explained_variance_ratios_values = [1.0, 0.0]
 
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(pca_array.shape, DATA.shape)
-
-    np.testing.assert_array_almost_equal(pca_array, expected_pca_values, decimal=3)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=3)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=3
+    _assert_expected_values(
+        pca_array=pca_array,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
+        expected_pca_values=expected_pca_values,
+        # Original implementation expected order as defined by EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE
+        expected_component_values=EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE,
+        expected_component_values_alternative=EXPECTED_DATA_COMPONENT_VALUES,
+        expected_explained_variance_ratios_values=expected_explained_variance_ratios_values,
+        decimal_accuracy=3,
+        data_shape=DATA_WITH_NAN.shape,
     )
 
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_with_nan_replace():
     """Test that PCA function gives correct output for Numpy array input that has NaN values and replace strategy."""
-    data = np.array([[1, 1], [2, np.nan], [3, 3]])
     pca_array, principal_components, explained_variances, explained_variance_ratios = compute_pca(
-        data, 2, nodata_handling="replace"
+        DATA_WITH_NAN, 2, nodata_handling="replace"
     )
 
-    expected_pca_values = np.array([[-1.73205, 1.11022e-16], [0, 0], [1.73205, 1.11022e-16]])
-    expected_component_values = np.array([[0.707, 0.707], [0.707, -0.707]])
-    expected_explained_variance_ratios_values = [1.0, 4.10865e-33]
-
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(pca_array.shape, DATA.shape)
-
-    np.testing.assert_array_almost_equal(pca_array, expected_pca_values, decimal=3)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=3)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=3
+    _assert_expected_values(
+        pca_array=pca_array,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
+        expected_pca_values=EXPECTED_DATA_PCA_VALUES,
+        expected_component_values=EXPECTED_DATA_COMPONENT_VALUES,
+        expected_component_values_alternative=EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE,
+        expected_explained_variance_ratios_values=EXPECTED_DATA_EXPLAINED_VARIANCE_RATIOS_VALUES,
+        decimal_accuracy=3,
+        data_shape=DATA_WITH_NAN.shape,
     )
 
 
 @pytest.mark.xfail(sys.platform == "win32", reason="Results deviate on Windows.", raises=AssertionError)
 def test_pca_with_nodata_removal():
     """Test that PCA function gives correct output for input that has specified nodata values and removal strategy."""
-    data = np.array([[1, 1], [2, -9999], [3, 3]])
     pca_array, principal_components, explained_variances, explained_variance_ratios = compute_pca(
-        data, 2, nodata_handling="remove", nodata=-9999
+        DATA_WITH_NODATA, 2, nodata_handling="remove", nodata=-9999
     )
 
     expected_pca_values = np.array([[-1.414, 0.0], [np.nan, np.nan], [1.414, 0.0]])
-    expected_component_values = np.array([[0.707, 0.707], [-0.707, 0.707]])
     expected_explained_variance_ratios_values = [1.0, 0.0]
 
-    np.testing.assert_equal(principal_components.size, 4)
-    np.testing.assert_equal(explained_variances.size, 2)
-    np.testing.assert_equal(explained_variance_ratios.size, 2)
-    np.testing.assert_equal(pca_array.shape, DATA.shape)
-
-    np.testing.assert_array_almost_equal(pca_array, expected_pca_values, decimal=3)
-    np.testing.assert_array_almost_equal(principal_components, expected_component_values, decimal=3)
-    np.testing.assert_array_almost_equal(
-        explained_variance_ratios, expected_explained_variance_ratios_values, decimal=3
+    _assert_expected_values(
+        pca_array=pca_array,
+        principal_components=principal_components,
+        explained_variances=explained_variances,
+        explained_variance_ratios=explained_variance_ratios,
+        expected_pca_values=expected_pca_values,
+        # Original implementation expected order as defined by EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE
+        expected_component_values=EXPECTED_DATA_COMPONENT_VALUES_ALTERNATIVE,
+        expected_component_values_alternative=EXPECTED_DATA_COMPONENT_VALUES,
+        expected_explained_variance_ratios_values=expected_explained_variance_ratios_values,
+        decimal_accuracy=3,
+        data_shape=DATA_WITH_NODATA.shape,
     )