18898: Adds 'time_feature_is_universal' to infer_feature_attributes…

…, MINOR (#98)

Adds a boolean flag 'time_feature_is_universal' to
`infer_feature_attributes`.

If 'time_feature_is_universal' is specified by the user, then that value
is directly copied into the feature attributes of the specified time
feature under `features[time_feature_name]['time_series']['universal']`.

If the value is not specified, then nothing is stored there, and it's
left to the Engine to assume default behavior.
Default value is None.

howso/utilities/feature_attributes/infer_feature_attributes.py

@@ -213,6 +213,14 @@ def infer_feature_attributes(data: Union[pd.DataFrame, SQLRelationalDatastorePro
         This will cause the bounds for the start and end times set
         to the same bounds as observed in the original data.
 
+    time_feature_is_universal : bool, optional
+        If True, the time feature will be treated as universal and future data
+        is excluded while making predictions. If False, the time feature will
+        not be treated as universal and only future data within the same series
+        is excluded while making predictions. It is recommended to set this
+        value to True if there is any possibility of global relevancy of time,
+        which is the default behavior.
+
     time_series_type_default : str, default 'rate'
         (Optional) Type specifying how time series is generated.
         One of 'rate' or 'delta', default is 'rate'. If 'rate',

howso/utilities/feature_attributes/tests/test_infer_time_series_attributes.py

@@ -199,3 +199,32 @@ def test_set_rate_delta_boundaries():
         assert 'rate_max' not in features['date']['time_series']
         assert 'delta_min' not in features['f3']['time_series']
         assert 'delta_max' not in features['f3']['time_series']
+
+
+@pytest.mark.parametrize(
+    ("universal_value", "expected"),
+    [
+        (True, True),
+        (False, False),
+        (None, None),
+    ]
+)
+def test_time_feature_is_universal(universal_value, expected):
+    """Validates that time_feature_is_universal is working as expected."""
+    df = pd.read_csv(data_path)
+
+    # Define time format
+    time_format = "%Y%m%d"
+    # Identify id-feature and time-feature
+    id_feature_name = "ID"
+    time_feature_name = "date"
+
+    features = infer_feature_attributes(
+        df,
+        time_feature_name=time_feature_name,
+        id_feature_name=id_feature_name,
+        datetime_feature_formats={time_feature_name: time_format},
+        time_feature_is_universal=universal_value,
+    )
+
+    assert features[time_feature_name]['time_series'].get("universal") == expected

howso/utilities/feature_attributes/time_series.py

@@ -259,6 +259,7 @@ def _process(  # noqa: C901
         attempt_infer_extended_nominals: bool = False,
         nominal_substitution_config: Optional[Dict[str, Dict]] = None,
         include_extended_nominal_probabilities: Optional[bool] = False,
+        time_feature_is_universal: bool = None,
         time_series_type_default: Optional[str] = 'rate',
         time_series_types_override: Optional[Dict] = None,
         orders_of_derivatives: Optional[Dict] = None,
@@ -423,6 +424,14 @@ def _process(  # noqa: C901
             (Optional) If true, extended nominal probabilities will be appended
             as metadata into the feature object.
 
+        time_feature_is_universal : bool, optional
+            If True, the time feature will be treated as universal and future data
+            is excluded while making predictions. If False, the time feature will
+            not be treated as universal and only future data within the same series
+            is excluded while making predictions. It is recommended to set this
+            value to True if there is any possibility of global relevancy of time,
+            which is the default behavior.
+
         time_series_type_default : str, default 'rate'
             (Optional) Type specifying how time series is generated.
             One of 'rate' or 'delta', default is 'rate'. If 'rate',
@@ -616,6 +625,10 @@ def _process(  # noqa: C901
 
         if self.time_feature_name in features:
             features[self.time_feature_name]['time_series']['time_feature'] = True
+
+            # Assign universal value if specified
+            if time_feature_is_universal is not None:
+                features[self.time_feature_name]['time_series']['universal'] = time_feature_is_universal
             # Force time_feature to be `continuous`
             features[self.time_feature_name]['type'] = "continuous"
             # Set time_series as 'delta' so that lag and delta are computed