sdmetrics 0.29.1.dev0__tar.gz → 0.30.1.dev0__tar.gz

{sdmetrics-0.29.1.dev0/sdmetrics.egg-info → sdmetrics-0.30.1.dev0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sdmetrics
-Version: 0.29.1.dev0
+Version: 0.30.1.dev0
 Summary: Metrics for Synthetic Data Generation Projects
 Author-email: MIT Data To AI Lab <dailabmit@gmail.com>
 License-Expression: MIT

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/pyproject.toml

@@ -148,7 +148,7 @@ convention = 'google'
 add-ignore = ['D107', 'D407', 'D417']
 
 [tool.bumpversion]
-current_version = "0.29.1.dev0"
+current_version = "0.30.1.dev0"
 parse = '(?P<major>\d+)\.(?P<minor>\d+)\.(?P<patch>\d+)(\.(?P<release>[a-z]+)(?P<candidate>\d+))?'
 serialize = [
     '{major}.{minor}.{patch}.{release}{candidate}',

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/__init__.py

@@ -4,7 +4,7 @@
 
 __author__ = 'MIT Data To AI Lab'
 __email__ = 'dailabmit@gmail.com'
-__version__ = '0.29.1.dev0'
+__version__ = '0.30.1.dev0'
 
 import sys
 import warnings as python_warnings

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/_utils_metadata.py

@@ -4,7 +4,7 @@ import pandas as pd
 
 from sdmetrics.utils import get_columns_from_metadata, is_datetime
 
-MODELABLE_SDTYPES = ('numerical', 'datetime', 'categorical', 'boolean')
+MODELABLE_SDTYPES = ('numerical', 'datetime', 'categorical', 'boolean', 'ordinal')
 
 
 def _validate_metadata_dict(metadata):
@@ -173,7 +173,7 @@ def _remove_missing_columns_metadata(data, metadata):
 def _remove_non_modelable_columns(data, metadata):
     """Remove columns that are not modelable.
 
-    All modelable columns are numerical, datetime, categorical, or boolean sdtypes.
+    All modelable columns are numerical, datetime, categorical, ordinal, or boolean sdtypes.
     """
     columns_modelable = []
     for column, column_metadata in get_columns_from_metadata(metadata).items():

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/multi_table/_properties/column_shapes.py

@@ -10,7 +10,7 @@ class ColumnShapes(BaseMultiTableProperty):
     This property assesses the shape similarity between the real and synthetic data.
     A metric score is computed column-wise and the final score is the average over all columns.
     The KSComplement metric is used for numerical and datetime columns while the TVComplement
-    is used for categorical and boolean columns.
+    is used for categorical, ordinal and boolean columns.
     The other column types are ignored by this property.
     """
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/multi_table/_properties/coverage.py

@@ -10,7 +10,7 @@ class Coverage(BaseMultiTableProperty):
     This property assesses data coverage between the real and synthetic data.
     A metric score is computed column-wise and the final score is the average over all columns.
     The ``RangeCoverage`` metric is used for numerical and datetime columns while the
-    ``CategoryCoverage`` is used for categorical and boolean columns.
+    ``CategoryCoverage`` is used for categorical, ordinal and boolean columns.
     The other column types are ignored by this property.
     """
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/multi_table/_properties/data_validity.py

@@ -11,7 +11,7 @@ class DataValidity(BaseMultiTableProperty):
     The metric is based on the type data in each column.
     A metric score is computed column-wise and the final score is the average over all columns.
     The BoundaryAdherence metric is used for numerical and datetime columns, the CategoryAdherence
-    is used for categorical and boolean columns and the KeyUniqueness for primary and
+    is used for categorical, ordinal and boolean columns and the KeyUniqueness for primary and
     alternate keys. The other column types are ignored by this property.
     """
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/single_table/_properties/column_pair_trends.py

@@ -25,6 +25,7 @@ class ColumnPairTrends(BaseSingleTableProperty):
         'numerical': 'continuous',
         'datetime': 'continuous',
         'categorical': 'discrete',
+        'ordinal': 'discrete',
         'boolean': 'discrete',
     }
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/single_table/_properties/column_shapes.py

@@ -14,7 +14,7 @@ class ColumnShapes(BaseSingleTableProperty):
     This property assesses the shape similarity between the real and synthetic data.
     A metric score is computed column-wise and the final score is the average over all columns.
     The KSComplement metric is used for numerical and datetime columns while the TVComplement
-    is used for categorical and boolean columns.
+    is used for categorical, ordinal and boolean columns.
     The other column types are ignored by this property.
     """
 
@@ -23,6 +23,7 @@ class ColumnShapes(BaseSingleTableProperty):
         'numerical': KSComplement,
         'datetime': KSComplement,
         'categorical': TVComplement,
+        'ordinal': TVComplement,
         'boolean': TVComplement,
     }
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/single_table/_properties/coverage.py

@@ -13,7 +13,7 @@ class Coverage(BaseSingleTableProperty):
     This property assesses data coverage between the real and synthetic data.
     A metric score is computed column-wise and the final score is the average over all columns.
     The ``RangeCoverage`` metric is used for numerical and datetime columns while the
-    ``CategoryCoverage`` is used for categorical and boolean columns.
+    ``CategoryCoverage`` is used for categorical, ordinal and boolean columns.
     The other column types are ignored by this property.
     """
 
@@ -22,6 +22,7 @@ class Coverage(BaseSingleTableProperty):
         'numerical': RangeCoverage,
         'datetime': RangeCoverage,
         'categorical': CategoryCoverage,
+        'ordinal': CategoryCoverage,
         'boolean': CategoryCoverage,
     }
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/single_table/_properties/data_validity.py

@@ -26,7 +26,7 @@ class DataValidity(BaseSingleTableProperty):
     This property computes, at base, whether each column contains valid data.
     The metric is based on the type data in each column.
     The BoundaryAdherence metric is used for numerical and datetime columns, the CategoryAdherence
-    is used for categorical and boolean columns and the KeyUniqueness for primary
+    is used for categorical, ordinal and boolean columns and the KeyUniqueness for primary
     and alternate keys. The other column types are ignored by this property.
 
     If the metadata defines the valid range of a column, it is passed down to the metric,
@@ -40,10 +40,15 @@ class DataValidity(BaseSingleTableProperty):
         BoundaryAdherence: ('range_min', 'range_max', 'range_is_nullable'),
         CategoryAdherence: ('range_values', 'range_is_nullable'),
     }
+    _metric_to_required_argument = {
+        DatetimeFormatAdherence: 'datetime_format',
+        RegexFormatAdherence: 'regex_format',
+    }
     _sdtype_to_metric = {
         'numerical': [BoundaryAdherence],
         'datetime': [BoundaryAdherence, DatetimeFormatAdherence],
         'categorical': [CategoryAdherence],
+        'ordinal': [CategoryAdherence],
         'boolean': [CategoryAdherence],
         'id': [KeyUniqueness, RegexFormatAdherence],
     }
@@ -75,6 +80,39 @@ class DataValidity(BaseSingleTableProperty):
         }
         return column_arguments
 
+    @classmethod
+    def _has_required_argument(cls, metric, column_name, columns_meta):
+        required_argument = cls._metric_to_required_argument.get(metric)
+        return required_argument is None or required_argument in columns_meta[column_name]
+
+    @classmethod
+    def _get_column_metrics(cls, sdtype, column_name, columns_meta, is_unique):
+        """Get the metrics that apply to a column.
+
+        ``KeyUniqueness`` only applies to primary and alternate keys, and the format metrics
+        only apply to the columns that define their format in the metadata.
+
+        Args:
+            sdtype (str or None):
+                The sdtype of the column, or ``None``.
+            column_name (str or list):
+                The name of the column.
+            columns_meta (dict):
+                The metadata of every column of the table.
+            is_unique (bool):
+                Whether the column is a primary or an alternate key.
+
+        Returns:
+            list:
+                The metrics to compute the column scores with.
+        """
+        return [
+            metric
+            for metric in cls._sdtype_to_metric.get(sdtype, [KeyUniqueness])
+            if (is_unique or metric is not KeyUniqueness)
+            and cls._has_required_argument(metric, column_name, columns_meta)
+        ]
+
     def _generate_details(self, real_data, synthetic_data, metadata, progress_bar=None):
         """Generate the _details dataframe for the data validity property.
 
@@ -110,13 +148,10 @@ class DataValidity(BaseSingleTableProperty):
             is_unique = primary_key_match or alternate_key_match
             is_sequence_index = column_name == sequence_index
 
-            if (sdtype == 'id' or sdtype not in self._sdtype_to_metric) and not is_unique:
-                continue
+            metrics = self._get_column_metrics(sdtype, column_name, columns_meta, is_unique)
+            if is_sequence_index and BoundaryAdherence in metrics:
+                metrics = []
 
-            if is_sequence_index and BoundaryAdherence in self._sdtype_to_metric.get(sdtype):
-                continue
-
-            metrics = self._sdtype_to_metric.get(sdtype, [KeyUniqueness])
             for metric in metrics:
                 try:
                     metric_arguments = self._get_metric_arguments(metric, column_name, columns_meta)
@@ -141,15 +176,15 @@ class DataValidity(BaseSingleTableProperty):
                 except Exception as e:
                     column_score = np.nan
                     error_message = f'{type(e).__name__}: {e}'
-                finally:
-                    if progress_bar:
-                        progress_bar.update()
 
                 column_names.append(column_name)
                 metric_names.append(metric.__name__)
                 scores.append(column_score)
                 error_messages.append(error_message)
 
+            if progress_bar:
+                progress_bar.update()
+
         result = pd.DataFrame({
             'Column': column_names,
             'Metric': metric_names,

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/reports/utils.py

@@ -16,7 +16,7 @@ from sdmetrics.utils import (
 )
 
 CONTINUOUS_SDTYPES = ['numerical', 'datetime']
-DISCRETE_SDTYPES = ['categorical', 'boolean']
+DISCRETE_SDTYPES = ['categorical', 'ordinal', 'boolean']
 DEFAULT_NUM_ROWS_SUBSAMPLE = 50000
 DEFAULT_REAL_CORRELATION_THRESHOLD = 0.5
 DEFAULT_REAL_ASSOCIATION_THRESHOLD = 0.3
@@ -99,7 +99,7 @@ def discretize_table_data(real_data, synthetic_data, metadata):
 
 
 def _get_non_id_columns(metadata, binned_metadata):
-    valid_sdtypes = ['numerical', 'categorical', 'boolean', 'datetime']
+    valid_sdtypes = ['numerical', 'categorical', 'boolean', 'ordinal', 'datetime']
     alternate_keys = get_alternate_keys(metadata)
     non_id_columns = []
     for column, column_meta in get_columns_from_metadata(binned_metadata).items():
@@ -202,7 +202,7 @@ def _validate_categorical_values(real_data, synthetic_data, metadata, table=None
     columns = get_columns_from_metadata(metadata)
     for column, column_meta in columns.items():
         column_type = get_type_from_column_meta(column_meta)
-        if column_type == 'categorical':
+        if column_type in ('categorical', 'ordinal'):
             extra_categories = [
                 value
                 for value in synthetic_data[column].unique()

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/single_column/statistical/regex_format_adherence.py

@@ -51,7 +51,10 @@ class RegexFormatAdherence(SingleColumnMetric):
             regex_format (str):
                 The regex format.
             compare (pd.Series | pd.DataFrame, optional):
-                Dataframe containing groups to compare against.
+                Dataframe containing groups to compare against. Only the named groups are
+                compared, and each one of them separately, so a value is valid when every
+                one of its named groups is present in the matching group of ``compare``,
+                even if the combination is not.
 
         Returns:
             tuple (pd.Series, pd.DataFrame):
@@ -59,13 +62,18 @@ class RegexFormatAdherence(SingleColumnMetric):
                 * A dataframe containing seperated regex groups.
         """
         regex_column = column.str.fullmatch(regex_format, na=False)
+        pattern = re.compile(regex_format)
 
         groups = None
-        if re.compile(regex_format).groups:
+        if pattern.groups:
             groups = column[regex_column].str.extract(regex_format)
-            if compare is not None and not compare.empty:
-                valid_groups = set(map(tuple, compare.dropna().to_numpy()))
-                regex_column = groups.apply(tuple, axis=1).isin(valid_groups)
+            if pattern.groupindex and compare is not None and not compare.empty:
+                valid_values = {
+                    group: set(compare.iloc[:, group_number - 1])
+                    for group, group_number in pattern.groupindex.items()
+                }
+                valid_groups = groups[list(valid_values)].isin(valid_values).all(axis=1)
+                regex_column &= valid_groups.reindex(column.index, fill_value=False)
 
         valid = pd.isna(column) | regex_column
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0}/sdmetrics/single_table/new_row_synthesis.py

@@ -101,7 +101,7 @@ class NewRowSynthesis(SingleTableMetric):
             numerical_fields = []
 
         try:
-            categorical_fields = cls._select_fields(metadata, ('categorical', 'boolean'))
+            categorical_fields = cls._select_fields(metadata, ('categorical', 'boolean', 'ordinal'))
         except IncomputableMetricError:
             categorical_fields = []
 

{sdmetrics-0.29.1.dev0 → sdmetrics-0.30.1.dev0/sdmetrics.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sdmetrics
-Version: 0.29.1.dev0
+Version: 0.30.1.dev0
 Summary: Metrics for Synthetic Data Generation Projects
 Author-email: MIT Data To AI Lab <dailabmit@gmail.com>
 License-Expression: MIT