Momo lookback windown/offset (Azure#33462)

* lookback window/offset

* e2e tests

* e2e tests

* pylint

sdk/ml/azure-ai-ml/assets.json

@@ -2,5 +2,5 @@
   "AssetsRepo": "Azure/azure-sdk-assets",
   "AssetsRepoPrefixPath": "python",
   "TagPrefix": "python/ml/azure-ai-ml",
-  "Tag": "python/ml/azure-ai-ml_9bd27ef67e"
+  "Tag": "python/ml/azure-ai-ml_f133080b91"
 }

sdk/ml/azure-ai-ml/azure/ai/ml/_schema/monitoring/signals.py

@@ -60,8 +60,8 @@ def make(self, data, **kwargs):
 class BaselineDataRangeSchema(metaclass=PatchedSchemaMeta):
     window_start = fields.Str()
     window_end = fields.Str()
-    trailing_window_size = fields.Str()
-    trailing_window_offset = fields.Str()
+    lookback_window_size = fields.Str()
+    lookback_window_offset = fields.Str()
 
     @post_load
     def make(self, data, **kwargs):
@@ -74,7 +74,7 @@ class ProductionDataSchema(metaclass=PatchedSchemaMeta):
     input_data = UnionField(union_fields=[NestedField(DataInputSchema), NestedField(MLTableInputSchema)])
     data_context = StringTransformedEnum(allowed_values=[o.value for o in MonitorDatasetContext])
     pre_processing_component = fields.Str()
-    data_window_size = fields.Str()
+    data_window = NestedField(BaselineDataRangeSchema)
 
     @post_load
     def make(self, data, **kwargs):
@@ -187,7 +187,7 @@ class FADProductionDataSchema(metaclass=PatchedSchemaMeta):
         keys=StringTransformedEnum(allowed_values=[o.value for o in FADColumnNames]), values=fields.Str()
     )
     pre_processing_component = fields.Str()
-    data_window_size = fields.Str()
+    data_window = NestedField(BaselineDataRangeSchema)
 
     @post_load
     def make(self, data, **kwargs):
@@ -283,7 +283,7 @@ def make(self, data, **kwargs):
 class LlmDataSchema(metaclass=PatchedSchemaMeta):
     input_data = UnionField(union_fields=[NestedField(DataInputSchema), NestedField(MLTableInputSchema)])
     data_column_names = fields.Dict()
-    data_window_size = fields.Str()
+    data_window = NestedField(BaselineDataRangeSchema)
 
     @post_load
     def make(self, data, **kwargs):

sdk/ml/azure-ai-ml/azure/ai/ml/operations/_schedule_operations.py

@@ -370,7 +370,9 @@ def _resolve_monitor_schedule_arm_id(  # pylint:disable=too-many-branches,too-ma
                                     type=self._data_operations.get(model_inputs_name, model_inputs_version).type,
                                 ),
                                 data_context=MonitorDatasetContext.MODEL_INPUTS,
-                                data_window_size="P7D",
+                                data_window=BaselineDataRange(
+                                    lookback_window_size="default", lookback_window_offset="P0D"
+                                ),
                             )
                         if not signal.reference_data:
                             signal.reference_data = ReferenceData(
@@ -379,7 +381,9 @@ def _resolve_monitor_schedule_arm_id(  # pylint:disable=too-many-branches,too-ma
                                     type=self._data_operations.get(model_inputs_name, model_inputs_version).type,
                                 ),
                                 data_context=MonitorDatasetContext.MODEL_INPUTS,
-                                data_window=BaselineDataRange(trailing_window_size="P7D", trailing_window_offset="P7D"),
+                                data_window=BaselineDataRange(
+                                    lookback_window_size="P7D", lookback_window_offset="default"
+                                ),
                             )
                     elif not mdc_input_enabled and not (signal.production_data and signal.reference_data):
                         # if target or baseline dataset is absent and data collector for input is not enabled,
@@ -401,7 +405,9 @@ def _resolve_monitor_schedule_arm_id(  # pylint:disable=too-many-branches,too-ma
                                     type=self._data_operations.get(model_outputs_name, model_outputs_version).type,
                                 ),
                                 data_context=MonitorDatasetContext.MODEL_OUTPUTS,
-                                data_window_size="P7D",
+                                data_window=BaselineDataRange(
+                                    lookback_window_size="default", lookback_window_offset="P0D"
+                                ),
                             )
                         if not signal.reference_data:
                             signal.reference_data = ReferenceData(
@@ -410,7 +416,9 @@ def _resolve_monitor_schedule_arm_id(  # pylint:disable=too-many-branches,too-ma
                                     type=self._data_operations.get(model_outputs_name, model_outputs_version).type,
                                 ),
                                 data_context=MonitorDatasetContext.MODEL_OUTPUTS,
-                                data_window=BaselineDataRange(trailing_window_size="P7D", trailing_window_offset="P7D"),
+                                data_window=BaselineDataRange(
+                                    lookback_window_size="P7D", lookback_window_offset="default"
+                                ),
                             )
                     elif not mdc_output_enabled and not (signal.production_data and signal.reference_data):
                         # if target dataset is absent and data collector for output is not enabled,
@@ -433,15 +441,19 @@ def _resolve_monitor_schedule_arm_id(  # pylint:disable=too-many-branches,too-ma
                                         type=self._data_operations.get(model_inputs_name, model_inputs_version).type,
                                     ),
                                     data_context=MonitorDatasetContext.MODEL_INPUTS,
-                                    data_window_size="P7D",
+                                    data_window=BaselineDataRange(
+                                        lookback_window_size="default", lookback_window_offset="P0D"
+                                    ),
                                 ),
                                 FADProductionData(
                                     input_data=Input(
                                         path=f"{model_outputs_name}:{model_outputs_version}",
                                         type=self._data_operations.get(model_outputs_name, model_outputs_version).type,
                                     ),
                                     data_context=MonitorDatasetContext.MODEL_OUTPUTS,
-                                    data_window_size="P7D",
+                                    data_window=BaselineDataRange(
+                                        lookback_window_size="default", lookback_window_offset="P0D"
+                                    ),
                                 ),
                             ]
                     elif not mdc_output_enabled and not signal.production_data:

sdk/ml/azure-ai-ml/tests/monitoring/unittests/test_monitor_schedule.py

@@ -87,10 +87,10 @@ def test_default_data_window_size_recurrence(self, test_path) -> None:
 
         # null out lookback
         for signal in schedule.create_monitor.monitoring_signals.values():
-            signal.production_data.data_window_size = None
+            signal.production_data.data_window.lookback_window_size = None
 
         # test minute
-        override_frequency_interval_and_check_window_size(schedule, "minute", 1, 1)
+        override_frequency_interval_and_check_window_size(schedule, "minute", 5, 1)
         override_frequency_interval_and_check_window_size(schedule, "hour", 5, 1)
         override_frequency_interval_and_check_window_size(schedule, "day", 6, 6)
         override_frequency_interval_and_check_window_size(schedule, "week", 2, 14)
@@ -102,8 +102,9 @@ def override_frequency_interval_and_check_window_size(
 ):
     schedule.trigger.frequency = frequency
     schedule.trigger.interval = interval
+
     for signal in schedule.create_monitor.monitoring_signals.values():
-        signal.production_data.data_window_size = None
+        signal.production_data.data_window.lookback_window_size = None
 
     to_rest_schedule = schedule._to_rest_object()
     for signal in to_rest_schedule.properties.action.monitor_definition.signals.values():

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/custom.yaml

@@ -48,8 +48,8 @@ create_monitor:
             path: azureml:Direct:1
           data_context: test
           data_window:
-            trailing_window_size: P30D
-            trailing_window_offset: P7D
+            lookback_window_size: P30D
+            lookback_window_offset: P7D
         test_static:
           input_data:
             type: mltable

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/data_drift.yaml

@@ -33,15 +33,17 @@ create_monitor:
           type: mltable
           path: azureml:Direct:1
         data_context: test
-        data_window_size: P60D
+        data_window:
+          lookback_window_size: P60D
+          lookback_window_offset: P0D
       reference_data:
         input_data:
           type: mltable
           path: azureml:Direct:1
         data_context: test
         data_window: 
-          trailing_window_size: P30D
-          trailing_window_offset: P30D
+          lookback_window_size: P30D
+          lookback_window_offset: P30D
       metric_thresholds:
         numerical:
           jensen_shannon_distance: 0.5

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/data_quality.yaml

@@ -33,7 +33,9 @@ create_monitor:
           type: mltable
           path: azureml:Direct:1
         data_context: test
-        data_window_size: "P60D"
+        data_window:
+          lookback_window_size: P60D
+          lookback_window_offset: P0D
       reference_data:
         input_data:
           type: mltable

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/feature_attribution_drift.yaml

@@ -34,7 +34,9 @@ create_monitor:
           data_context: model_inputs
           data_column_names:
             correlation_id: inference_correlation_id
-          data_window_size: P7D
+          data_window:
+            lookback_window_size: P7D
+            lookback_window_offset: P0D
         - input_data:
             path: azureml:my_model_outputs_data:1
             type: mltable
@@ -43,7 +45,9 @@ create_monitor:
             prediction: is_fraud
             prediction_probability: is_fraud_probability
             correlation_id: inference_correlation_id
-          data_window_size: P7D
+          data_window:
+            lookback_window_size: P7D
+            lookback_window_offset: P0D
       reference_data:
         input_data:
           path: azureml:my_model_training_data:1

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/generation_safety.yaml

@@ -32,13 +32,17 @@ create_monitor:
         - input_data:
             path: azureml:my_model_inputs_data:1
             type: mltable
-          data_window_size: P7D
+          data_window:
+            lookback_window_size: P7D
+            lookback_window_offset: P0D
           data_column_names:
             correlation_id: inference_correlation_id
         - input_data:
             path: azureml:my_model_outputs_data:1
             type: mltable
-          data_window_size: P7D
+          data_window:
+            lookback_window_size: P7D
+            lookback_window_offset: P0D
           data_column_names:
             prediction: is_fraud
             prediction_probability: is_fraud_probability

sdk/ml/azure-ai-ml/tests/test_configs/monitoring/yaml_configs/prediction_drift.yaml

@@ -33,7 +33,9 @@ create_monitor:
             type: mltable
             path: azureml:Direct:1
         data_context: test
-        data_window_size: "P60D"
+        data_window:
+          lookback_window_size: P60D
+          lookback_window_offset: P0D
       reference_data:
         input_data:
           type: mltable