Fix KSComplement instability for constant float values (#654)

Author: fealho

sdmetrics/single_column/statistical/kscomplement.py

@@ -1,5 +1,7 @@
 """Kolmogorov-Smirnov test based Metric."""
 
+import sys
+
 import numpy as np
 import pandas as pd
 from scipy.stats import ks_2samp
@@ -8,6 +10,8 @@
 from sdmetrics.single_column.base import SingleColumnMetric
 from sdmetrics.utils import is_datetime
 
+MAX_DECIMALS = sys.float_info.dig - 1
+
 
 class KSComplement(SingleColumnMetric):
     """Kolmogorov-Smirnov statistic based metric.
@@ -57,6 +61,9 @@ def compute(real_data, synthetic_data):
             real_data = pd.to_numeric(real_data)
             synthetic_data = pd.to_numeric(synthetic_data)
 
+        real_data = real_data.round(MAX_DECIMALS)
+        synthetic_data = synthetic_data.round(MAX_DECIMALS)
+
         try:
             statistic, _ = ks_2samp(real_data, synthetic_data)
         except ValueError as e:

tests/integration/reports/multi_table/test_quality_report.py

@@ -3,6 +3,7 @@
 
 import numpy as np
 import pandas as pd
+from packaging import version
 
 from sdmetrics.demos import load_demo
 from sdmetrics.reports.multi_table.quality_report import QualityReport
@@ -299,6 +300,15 @@ def test_quality_report_with_errors():
         'Property': ['Column Shapes', 'Column Pair Trends', 'Cardinality', 'Intertable Trends'],
         'Score': [0.8165079365079364, 0.55, 0.95, 0.5833333333333334],
     })
+
+    pandas_version = version.parse(pd.__version__)
+    if pandas_version >= version.parse('2.2.0'):
+        err1 = "TypeError: '<' not supported between instances of 'int' and 'str'"
+        err2 = "TypeError: '<' not supported between instances of 'Timestamp' and 'str'"
+        err3 = "TypeError: '<' not supported between instances of 'float' and 'str'"
+    else:
+        err1 = err2 = err3 = "TypeError: can't multiply sequence by non-int of type 'float'"
+
     expected_details = pd.DataFrame({
         'Table': [
             'users',
@@ -334,11 +344,11 @@ def test_quality_report_with_errors():
         'Error': [
             None,
             None,
-            "TypeError: '<' not supported between instances of 'int' and 'str'",
+            err1,
             np.nan,
             np.nan,
-            "TypeError: '<' not supported between instances of 'Timestamp' and 'str'",
-            "TypeError: '<' not supported between instances of 'float' and 'str'",
+            err2,
+            err3,
             None,
         ],
     })

tests/integration/reports/single_table/_properties/test_column_shapes.py

@@ -1,4 +1,5 @@
 import pandas as pd
+from packaging import version
 
 from sdmetrics.demos import load_demo
 from sdmetrics.reports.single_table._properties import ColumnShapes
@@ -84,15 +85,22 @@ def test_get_score_errors(self):
 
         # Run
         column_shape_property = ColumnShapes()
-
-        expected_message_1 = (
-            "TypeError: '<' not supported between instances of 'Timestamp' and 'int'"
-        )
-        expected_message_2 = "TypeError: '<' not supported between instances of 'str' and 'float'"
-
         score = column_shape_property.get_score(real_data, synthetic_data, metadata)
 
         # Assert
+        pandas_version = version.parse(pd.__version__)
+        if pandas_version >= version.parse('2.2.0'):
+            expected_message_1 = (
+                "TypeError: '<' not supported between instances of 'Timestamp' and 'int'"
+            )
+            expected_message_2 = (
+                "TypeError: '<' not supported between instances of 'str' and 'float'"
+            )
+        else:
+            expected_message_1 = (
+                "TypeError: unsupported operand type(s) for *: 'Timestamp' and 'float'"
+            )
+            expected_message_2 = "TypeError: can't multiply sequence by non-int of type 'float'"
 
         details = column_shape_property.details
         details_nan = details.loc[pd.isna(details['Score'])]

tests/integration/single_column/statistical/test_kscomplement.py

@@ -46,3 +46,16 @@ def test_bad(array_like):
 
     assert 0.0 <= output < 0.5
     assert 0.0 <= normalized < 0.5
+
+
+def test_one_float_value():
+    """Test KSComplement.compute when both data have the same float values GH#652."""
+    # Setup
+    real = pd.Series([0.3 - 0.2])
+    synth = pd.Series([0.2 - 0.1])
+
+    # Run
+    output = KSComplement.compute(real, synth)
+
+    # Assert
+    assert output == 1

tests/unit/reports/single_table/_properties/test_column_shapes.py

@@ -2,6 +2,7 @@
 
 import numpy as np
 import pandas as pd
+from packaging import version
 
 from sdmetrics.reports.single_table._properties.column_shapes import ColumnShapes
 
@@ -108,7 +109,12 @@ def test__generate_details_error(self):
         result = column_shape_property._generate_details(real_data, synthetic_data, metadata)
 
         # Assert
-        expected_message = "TypeError: '<' not supported between instances of 'str' and 'int'"
+        pandas_version = version.parse(pd.__version__)
+        if pandas_version >= version.parse('2.2.0'):
+            expected_message = "TypeError: '<' not supported between instances of 'str' and 'int'"
+        else:
+            expected_message = "TypeError: can't multiply sequence by non-int of type 'float'"
+
         result_nan = result.loc[pd.isna(result['Score'])]
         column_names_nan = result_nan['Column'].tolist()
         error_message = result_nan['Error'].tolist()