workbench 0.8.160__py3-none-any.whl → 0.8.202__py3-none-any.whl

workbench/core/transforms/features_to_model/features_to_model.py

@@ -9,7 +9,7 @@ import time
 # Local Imports
 from workbench.core.transforms.transform import Transform, TransformInput, TransformOutput
 from workbench.core.artifacts.feature_set_core import FeatureSetCore
-from workbench.core.artifacts.model_core import ModelCore, ModelType, ModelImages
+from workbench.core.artifacts.model_core import ModelCore, ModelType, ModelFramework, ModelImages
 from workbench.core.artifacts.artifact import Artifact
 from workbench.model_scripts.script_generation import generate_model_script, fill_template
 from workbench.utils.model_utils import supported_instance_types
@@ -33,12 +33,13 @@ class FeaturesToModel(Transform):
         feature_name: str,
         model_name: str,
         model_type: ModelType,
+        model_framework=ModelFramework.XGBOOST,
         model_class=None,
         model_import_str=None,
         custom_script=None,
         custom_args=None,
-        training_image="xgb_training",
-        inference_image="xgb_inference",
+        training_image="training",
+        inference_image="inference",
         inference_arch="x86_64",
     ):
         """FeaturesToModel Initialization
@@ -46,12 +47,13 @@ class FeaturesToModel(Transform):
             feature_name (str): Name of the FeatureSet to use as input
             model_name (str): Name of the Model to create as output
             model_type (ModelType): ModelType.REGRESSOR or ModelType.CLASSIFIER, etc.
+            model_framework (ModelFramework, optional): The model framework (default ModelFramework.XGBOOST)
             model_class (str, optional): The scikit model (e.g. KNeighborsRegressor) (default None)
             model_import_str (str, optional): The import string for the model (default None)
             custom_script (str, optional): Custom script to use for the model (default None)
             custom_args (dict, optional): Custom arguments to pass to custom model scripts (default None)
-            training_image (str, optional): Training image (default "xgb_training")
-            inference_image (str, optional): Inference image (default "xgb_inference")
+            training_image (str, optional): Training image (default "training")
+            inference_image (str, optional): Inference image (default "inference")
             inference_arch (str, optional): Inference architecture (default "x86_64")
         """
 
@@ -65,6 +67,7 @@ class FeaturesToModel(Transform):
         self.input_type = TransformInput.FEATURE_SET
         self.output_type = TransformOutput.MODEL
         self.model_type = model_type
+        self.model_framework = model_framework
         self.model_class = model_class
         self.model_import_str = model_import_str
         self.custom_script = str(custom_script) if custom_script else None
@@ -157,6 +160,7 @@ class FeaturesToModel(Transform):
         template_params = {
             "model_imports": self.model_import_str,
             "model_type": self.model_type,
+            "model_framework": self.model_framework,
             "model_class": self.model_class,
             "target_column": self.target_column,
             "feature_list": self.model_feature_list,
@@ -164,6 +168,7 @@ class FeaturesToModel(Transform):
             "model_metrics_s3_path": self.model_training_root,
             "train_all_data": train_all_data,
             "id_column": feature_set.id_column,
+            "hyperparameters": kwargs.get("hyperparameters", {}),
         }
 
         # Custom Script
@@ -183,13 +188,15 @@ class FeaturesToModel(Transform):
             # Generate our model script
             script_path = generate_model_script(template_params)
 
-        # Metric Definitions for Regression
+        # Metric Definitions for Regression (matches model script output format)
         if self.model_type in [ModelType.REGRESSOR, ModelType.UQ_REGRESSOR, ModelType.ENSEMBLE_REGRESSOR]:
             metric_definitions = [
-                {"Name": "RMSE", "Regex": "RMSE: ([0-9.]+)"},
-                {"Name": "MAE", "Regex": "MAE: ([0-9.]+)"},
-                {"Name": "R2", "Regex": "R2: ([0-9.]+)"},
-                {"Name": "NumRows", "Regex": "NumRows: ([0-9]+)"},
+                {"Name": "rmse", "Regex": r"rmse: ([0-9.]+)"},
+                {"Name": "mae", "Regex": r"mae: ([0-9.]+)"},
+                {"Name": "medae", "Regex": r"medae: ([0-9.]+)"},
+                {"Name": "r2", "Regex": r"r2: ([0-9.-]+)"},
+                {"Name": "spearmanr", "Regex": r"spearmanr: ([0-9.-]+)"},
+                {"Name": "support", "Regex": r"support: ([0-9]+)"},
             ]
 
         # Metric Definitions for Classification
@@ -209,7 +216,7 @@ class FeaturesToModel(Transform):
                 raise ValueError(msg)
 
             # Dynamically create the metric definitions
-            metrics = ["precision", "recall", "fscore"]
+            metrics = ["precision", "recall", "f1"]
             metric_definitions = []
             for t in self.class_labels:
                 for m in metrics:
@@ -232,13 +239,21 @@ class FeaturesToModel(Transform):
         source_dir = str(Path(script_path).parent)
 
         # Create a Sagemaker Model with our script
-        image = ModelImages.get_image_uri(self.sm_session.boto_region_name, self.training_image, "0.1")
+        image = ModelImages.get_image_uri(self.sm_session.boto_region_name, self.training_image)
+
+        # Use GPU instance for ChemProp/PyTorch, CPU for others
+        if self.model_framework in [ModelFramework.CHEMPROP, ModelFramework.PYTORCH_TABULAR]:
+            train_instance_type = "ml.g6.xlarge"  # NVIDIA L4 GPU, ~$0.80/hr
+            self.log.important(f"Using GPU instance {train_instance_type} for {self.model_framework.value}")
+        else:
+            train_instance_type = "ml.m5.xlarge"
+
         self.estimator = Estimator(
             entry_point=entry_point,
             source_dir=source_dir,
             role=self.workbench_role_arn,
             instance_count=1,
-            instance_type="ml.m5.xlarge",
+            instance_type=train_instance_type,
             sagemaker_session=self.sm_session,
             image_uri=image,
             metric_definitions=metric_definitions,
@@ -263,13 +278,20 @@ class FeaturesToModel(Transform):
         self.log.important(f"Creating new model {self.output_name}...")
         self.create_and_register_model(**kwargs)
 
+        # Make a copy of the training view, to lock-in the training data used for this model
+        model_training_view_name = f"{self.output_name.replace('-', '_')}_training"
+        self.log.important(f"Creating Model Training View: {model_training_view_name}...")
+        feature_set.view("training").copy(f"{model_training_view_name}")
+
     def post_transform(self, **kwargs):
         """Post-Transform: Calling onboard() on the Model"""
         self.log.info("Post-Transform: Calling onboard() on the Model...")
         time.sleep(3)  # Give AWS time to complete Model register
 
-        # Store the model feature_list and target_column in the workbench_meta
-        output_model = ModelCore(self.output_name, model_type=self.model_type)
+        # Store the model metadata information
+        output_model = ModelCore(self.output_name)
+        output_model._set_model_type(self.model_type)
+        output_model._set_model_framework(self.model_framework)
         output_model.upsert_workbench_meta({"workbench_model_features": self.model_feature_list})
         output_model.upsert_workbench_meta({"workbench_model_target": self.target_column})
 
@@ -280,11 +302,12 @@ class FeaturesToModel(Transform):
         # Call the Model onboard method
         output_model.onboard_with_args(self.model_type, self.target_column, self.model_feature_list)
 
-    def create_and_register_model(self, aws_region=None):
+    def create_and_register_model(self, aws_region=None, **kwargs):
         """Create and Register the Model
 
         Args:
             aws_region (str, optional): AWS Region to use (default None)
+            **kwargs: Additional keyword arguments to pass to the model registration
         """
 
         # Get the metadata/tags to push into AWS
@@ -299,7 +322,7 @@ class FeaturesToModel(Transform):
 
         # Register our model
         image = ModelImages.get_image_uri(
-            self.sm_session.boto_region_name, self.inference_image, "0.1", self.inference_arch
+            self.sm_session.boto_region_name, self.inference_image, architecture=self.inference_arch
         )
         self.log.important(f"Registering model {self.output_name} with Inference Image {image}...")
         model = self.estimator.create_model(role=self.workbench_role_arn)
@@ -323,12 +346,11 @@ if __name__ == "__main__":
 
     # Regression Model
     input_name = "abalone_features"
-    output_name = "test-abalone-regression"
+    output_name = "abalone-regression"
     to_model = FeaturesToModel(input_name, output_name, model_type=ModelType.REGRESSOR)
     to_model.set_output_tags(["test"])
     to_model.transform(target_column="class_number_of_rings", description="Test Abalone Regression")
 
-    """
     # Classification Model
     input_name = "wine_features"
     output_name = "wine-classification"
@@ -338,10 +360,10 @@ if __name__ == "__main__":
 
     # Quantile Regression Model (Abalone)
     input_name = "abalone_features"
-    output_name = "abalone-quantile-reg"
+    output_name = "abalone-regression-uq"
     to_model = FeaturesToModel(input_name, output_name, ModelType.UQ_REGRESSOR)
-    to_model.set_output_tags(["abalone", "quantiles"])
-    to_model.transform(target_column="class_number_of_rings", description="Abalone Quantile Regression")
+    to_model.set_output_tags(["abalone", "uq"])
+    to_model.transform(target_column="class_number_of_rings", description="Abalone UQ Regression")
 
     # Scikit-Learn Kmeans Clustering Model
     input_name = "wine_features"
@@ -395,7 +417,7 @@ if __name__ == "__main__":
     scripts_root = Path(__file__).resolve().parents[3] / "model_scripts"
     my_script = scripts_root / "custom_models" / "chem_info" / "molecular_descriptors.py"
     input_name = "aqsol_features"
-    output_name = "smiles-to-taut-md-stereo-v0"
+    output_name = "test-smiles-to-taut-md-stereo"
     to_model = FeaturesToModel(input_name, output_name, model_type=ModelType.TRANSFORMER, custom_script=my_script)
     to_model.set_output_tags(["smiles", "molecular descriptors"])
     to_model.transform(target_column=None, feature_list=["smiles"], description="Smiles to Molecular Descriptors")
@@ -408,13 +430,3 @@ if __name__ == "__main__":
     to_model = FeaturesToModel(input_name, output_name, model_type=ModelType.TRANSFORMER, custom_script=my_script)
     to_model.set_output_tags(["smiles", "morgan fingerprints"])
     to_model.transform(target_column=None, feature_list=["smiles"], description="Smiles to Morgan Fingerprints")
-
-    # Tautomerization Model
-    scripts_root = Path(__file__).resolve().parents[3] / "model_scripts"
-    my_script = scripts_root / "custom_models" / "chem_info" / "tautomerize.py"
-    input_name = "aqsol_features"
-    output_name = "tautomerize-v0"
-    to_model = FeaturesToModel(input_name, output_name, model_type=ModelType.TRANSFORMER, custom_script=my_script)
-    to_model.set_output_tags(["smiles", "tautomerization"])
-    to_model.transform(target_column=None, feature_list=["smiles"], description="Tautomerize Smiles")
-    """

workbench/core/transforms/model_to_endpoint/model_to_endpoint.py

@@ -5,6 +5,7 @@ from sagemaker import ModelPackage
 from sagemaker.serializers import CSVSerializer
 from sagemaker.deserializers import CSVDeserializer
 from sagemaker.serverless import ServerlessInferenceConfig
+from sagemaker.model_monitor import DataCaptureConfig
 
 # Local Imports
 from workbench.core.transforms.transform import Transform, TransformInput, TransformOutput
@@ -51,27 +52,38 @@ class ModelToEndpoint(Transform):
         EndpointCore.managed_delete(self.output_name)
 
         # Get the Model Package ARN for our input model
-        input_model = ModelCore(self.input_name)
-        model_package_arn = input_model.model_package_arn()
+        workbench_model = ModelCore(self.input_name)
 
         # Deploy the model
-        self._deploy_model(model_package_arn, **kwargs)
+        self._deploy_model(workbench_model, **kwargs)
 
         # Add this endpoint to the set of registered endpoints for the model
-        input_model.register_endpoint(self.output_name)
+        workbench_model.register_endpoint(self.output_name)
 
         # This ensures that the endpoint is ready for use
         time.sleep(5)  # We wait for AWS Lag
         end = EndpointCore(self.output_name)
         self.log.important(f"Endpoint {end.name} is ready for use")
 
-    def _deploy_model(self, model_package_arn: str, mem_size: int = 2048, max_concurrency: int = 5):
+    def _deploy_model(
+        self,
+        workbench_model: ModelCore,
+        mem_size: int = 2048,
+        max_concurrency: int = 5,
+        data_capture: bool = False,
+        capture_percentage: int = 100,
+    ):
         """Internal Method: Deploy the Model
 
         Args:
-            model_package_arn(str): The Model Package ARN used to deploy the Endpoint
+            workbench_model(ModelCore): The Workbench ModelCore object to deploy
+            mem_size(int): Memory size for serverless deployment
+            max_concurrency(int): Max concurrency for serverless deployment
+            data_capture(bool): Enable data capture during deployment
+            capture_percentage(int): Percentage of data to capture. Defaults to 100.
         """
         # Grab the specified Model Package
+        model_package_arn = workbench_model.model_package_arn()
         model_package = ModelPackage(
             role=self.workbench_role_arn,
             model_package_arn=model_package_arn,
@@ -95,6 +107,23 @@ class ModelToEndpoint(Transform):
                 max_concurrency=max_concurrency,
             )
 
+        # Configure data capture if requested (and not serverless)
+        data_capture_config = None
+        if data_capture and not self.serverless:
+            # Set up the S3 path for data capture
+            base_endpoint_path = f"{workbench_model.endpoints_s3_path}/{self.output_name}"
+            data_capture_path = f"{base_endpoint_path}/data_capture"
+            self.log.important(f"Configuring Data Capture --> {data_capture_path}")
+            data_capture_config = DataCaptureConfig(
+                enable_capture=True,
+                sampling_percentage=capture_percentage,
+                destination_s3_uri=data_capture_path,
+            )
+        elif data_capture and self.serverless:
+            self.log.warning(
+                "Data capture is not supported for serverless endpoints. Skipping data capture configuration."
+            )
+
         # Deploy the Endpoint
         self.log.important(f"Deploying the Endpoint {self.output_name}...")
         model_package.deploy(
@@ -104,6 +133,7 @@ class ModelToEndpoint(Transform):
             endpoint_name=self.output_name,
             serializer=CSVSerializer(),
             deserializer=CSVDeserializer(),
+            data_capture_config=data_capture_config,
             tags=aws_tags,
         )
 

workbench/core/transforms/pandas_transforms/pandas_to_features.py

@@ -327,9 +327,36 @@ class PandasToFeatures(Transform):
         self.delete_existing()
         self.output_feature_group = self.create_feature_group()
 
+    def mac_spawn_hack(self):
+        """Workaround for macOS Tahoe fork/spawn issue with SageMaker FeatureStore ingest.
+
+        See: https://github.com/aws/sagemaker-python-sdk/issues/5312
+        macOS Tahoe 26+ has issues with forked processes creating boto3 sessions.
+        This forces spawn mode on macOS to avoid the hang.
+        """
+        import platform
+
+        if platform.system() == "Darwin":  # macOS
+            self.log.warning("macOS detected, forcing 'spawn' mode for multiprocessing (Tahoe hang workaround)")
+            import multiprocessing
+
+            try:
+                import multiprocess
+
+                multiprocess.set_start_method("spawn", force=True)
+            except (RuntimeError, ImportError):
+                pass  # Already set or multiprocess not available
+            try:
+                multiprocessing.set_start_method("spawn", force=True)
+            except RuntimeError:
+                pass  # Already set
+
     def transform_impl(self):
         """Transform Implementation: Ingest the data into the Feature Group"""
 
+        # Workaround for macOS Tahoe hang issue
+        self.mac_spawn_hack()
+
         # Now we actually push the data into the Feature Group (called ingestion)
         self.log.important(f"Ingesting rows into Feature Group {self.output_name}...")
         ingest_manager = self.output_feature_group.ingest(self.output_df, max_workers=8, max_processes=4, wait=False)

workbench/core/views/training_view.py

@@ -3,14 +3,18 @@
 from typing import Union
 
 # Workbench Imports
-from workbench.api import DataSource, FeatureSet
+from workbench.api import FeatureSet
 from workbench.core.views.view import View
 from workbench.core.views.create_view import CreateView
 from workbench.core.views.view_utils import get_column_list
 
 
 class TrainingView(CreateView):
-    """TrainingView Class: A View with an additional training column that marks holdout ids
+    """TrainingView Class: A View with an additional training column (80/20 or holdout ids).
+    The TrainingView class creates a SQL view that includes all columns from the source table
+    along with an additional boolean column named "training". This view can also include
+    a SQL filter expression to filter the rows included in the view.
+
 
     Common Usage:
         ```python
@@ -19,8 +23,9 @@ class TrainingView(CreateView):
         training_view = TrainingView.create(fs)
         df = training_view.pull_dataframe()
 
-        # Create a TrainingView with a specific set of columns
-        training_view = TrainingView.create(fs, column_list=["my_col1", "my_col2"])
+        # Create a TrainingView with a specific filter expression
+        training_view = TrainingView.create(fs, id_column="auto_id", filter_expression="age > 30")
+        df = training_view.pull_dataframe()
 
         # Query the view
         df = training_view.query(f"SELECT * FROM {training_view.table} where training = TRUE")
@@ -31,17 +36,21 @@ class TrainingView(CreateView):
     def create(
         cls,
         feature_set: FeatureSet,
-        source_table: str = None,
+        *,  # Enforce keyword arguments after feature_set
         id_column: str = None,
         holdout_ids: Union[list[str], list[int], None] = None,
+        filter_expression: str = None,
+        source_table: str = None,
     ) -> Union[View, None]:
         """Factory method to create and return a TrainingView instance.
 
         Args:
             feature_set (FeatureSet): A FeatureSet object
-            source_table (str, optional): The table/view to create the view from. Defaults to None.
             id_column (str, optional): The name of the id column. Defaults to None.
             holdout_ids (Union[list[str], list[int], None], optional): A list of holdout ids. Defaults to None.
+            filter_expression (str, optional): SQL filter expression (e.g., "age > 25 AND status = 'active'").
+                                               Defaults to None.
+            source_table (str, optional): The table/view to create the view from. Defaults to None.
 
         Returns:
             Union[View, None]: The created View object (or None if failed to create the view)
@@ -69,28 +78,36 @@ class TrainingView(CreateView):
                 else:
                     id_column = instance.auto_id_column
 
-        # If we don't have holdout ids, create a default training view
-        if not holdout_ids:
-            instance._default_training_view(instance.data_source, id_column)
-            return View(instance.data_source, instance.view_name, auto_create_view=False)
+        # Enclose each column name in double quotes
+        sql_columns = ", ".join([f'"{column}"' for column in column_list])
+
+        # Build the training assignment logic
+        if holdout_ids:
+            # Format the list of holdout ids for SQL IN clause
+            if all(isinstance(id, str) for id in holdout_ids):
+                formatted_holdout_ids = ", ".join(f"'{id}'" for id in holdout_ids)
+            else:
+                formatted_holdout_ids = ", ".join(map(str, holdout_ids))
 
-        # Format the list of holdout ids for SQL IN clause
-        if holdout_ids and all(isinstance(id, str) for id in holdout_ids):
-            formatted_holdout_ids = ", ".join(f"'{id}'" for id in holdout_ids)
+            training_logic = f"""CASE
+                WHEN {id_column} IN ({formatted_holdout_ids}) THEN False
+                ELSE True
+            END AS training"""
         else:
-            formatted_holdout_ids = ", ".join(map(str, holdout_ids))
+            # Default 80/20 split using modulo
+            training_logic = f"""CASE
+                WHEN MOD(ROW_NUMBER() OVER (ORDER BY {id_column}), 10) < 8 THEN True
+                ELSE False
+            END AS training"""
 
-        # Enclose each column name in double quotes
-        sql_columns = ", ".join([f'"{column}"' for column in column_list])
+        # Build WHERE clause if filter_expression is provided
+        where_clause = f"\nWHERE {filter_expression}" if filter_expression else ""
 
         # Construct the CREATE VIEW query
         create_view_query = f"""
         CREATE OR REPLACE VIEW {instance.table} AS
-        SELECT {sql_columns}, CASE
-            WHEN {id_column} IN ({formatted_holdout_ids}) THEN False
-            ELSE True
-        END AS training
-        FROM {instance.source_table}
+        SELECT {sql_columns}, {training_logic}
+        FROM {instance.source_table}{where_clause}
         """
 
         # Execute the CREATE VIEW query
@@ -99,35 +116,56 @@ class TrainingView(CreateView):
         # Return the View
         return View(instance.data_source, instance.view_name, auto_create_view=False)
 
-    # This is an internal method that's used to create a default training view
-    def _default_training_view(self, data_source: DataSource, id_column: str):
-        """Create a default view in Athena that assigns roughly 80% of the data to training
+    @classmethod
+    def create_with_sql(
+        cls,
+        feature_set: FeatureSet,
+        *,
+        sql_query: str,
+        id_column: str = None,
+    ) -> Union[View, None]:
+        """Factory method to create a TrainingView from a custom SQL query.
+
+        This method takes a complete SQL query and adds the default 80/20 training split.
+        Use this when you need complex queries like UNION ALL for oversampling.
 
         Args:
-            data_source (DataSource): The Workbench DataSource object
-            id_column (str): The name of the id column
+            feature_set (FeatureSet): A FeatureSet object
+            sql_query (str): Complete SELECT query (without the final semicolon)
+            id_column (str, optional): The name of the id column for training split. Defaults to None.
+
+        Returns:
+            Union[View, None]: The created View object (or None if failed)
         """
-        self.log.important(f"Creating default Training View {self.table}...")
+        # Instantiate the TrainingView
+        instance = cls("training", feature_set)
 
-        # Drop any columns generated from AWS
-        aws_cols = ["write_time", "api_invocation_time", "is_deleted", "event_time"]
-        column_list = [col for col in data_source.columns if col not in aws_cols]
+        # Sanity check on the id column
+        if not id_column:
+            instance.log.important("No id column specified, using auto_id_column")
+            if not instance.auto_id_column:
+                instance.log.error("No id column specified and no auto_id_column found, aborting")
+                return None
+            id_column = instance.auto_id_column
 
-        # Enclose each column name in double quotes
-        sql_columns = ", ".join([f'"{column}"' for column in column_list])
+        # Default 80/20 split using modulo
+        training_logic = f"""CASE
+            WHEN MOD(ROW_NUMBER() OVER (ORDER BY {id_column}), 10) < 8 THEN True
+            ELSE False
+        END AS training"""
 
-        # Construct the CREATE VIEW query with a simple modulo operation for the 80/20 split
+        # Wrap the custom query and add training column
         create_view_query = f"""
-        CREATE OR REPLACE VIEW "{self.table}" AS
-        SELECT {sql_columns}, CASE
-            WHEN MOD(ROW_NUMBER() OVER (ORDER BY {id_column}), 10) < 8 THEN True  -- Assign 80% to training
-            ELSE False  -- Assign roughly 20% to validation/test
-        END AS training
-        FROM {self.base_table_name}
+        CREATE OR REPLACE VIEW {instance.table} AS
+        SELECT *, {training_logic}
+        FROM ({sql_query}) AS custom_source
         """
 
         # Execute the CREATE VIEW query
-        data_source.execute_statement(create_view_query)
+        instance.data_source.execute_statement(create_view_query)
+
+        # Return the View
+        return View(instance.data_source, instance.view_name, auto_create_view=False)
 
 
 if __name__ == "__main__":
@@ -135,7 +173,7 @@ if __name__ == "__main__":
     from workbench.api import FeatureSet
 
     # Get the FeatureSet
-    fs = FeatureSet("test_features")
+    fs = FeatureSet("abalone_features")
 
     # Delete the existing training view
     training_view = TrainingView.create(fs)
@@ -152,9 +190,42 @@ if __name__ == "__main__":
 
     # Create a TrainingView with holdout ids
     my_holdout_ids = list(range(10))
-    training_view = TrainingView.create(fs, id_column="id", holdout_ids=my_holdout_ids)
+    training_view = TrainingView.create(fs, id_column="auto_id", holdout_ids=my_holdout_ids)
 
     # Pull the training data
     df = training_view.pull_dataframe()
     print(df.head())
     print(df["training"].value_counts())
+    print(f"Shape: {df.shape}")
+    print(f"Diameter min: {df['diameter'].min()}, max: {df['diameter'].max()}")
+
+    # Test the filter expression
+    training_view = TrainingView.create(fs, id_column="auto_id", filter_expression="diameter > 0.5")
+    df = training_view.pull_dataframe()
+    print(df.head())
+    print(f"Shape with filter: {df.shape}")
+    print(f"Diameter min: {df['diameter'].min()}, max: {df['diameter'].max()}")
+
+    # Test create_with_sql with a custom query (UNION ALL for oversampling)
+    print("\n--- Testing create_with_sql with oversampling ---")
+    base_table = fs.table
+    replicate_ids = [0, 1, 2]  # Oversample these IDs
+
+    custom_sql = f"""
+        SELECT * FROM {base_table}
+
+        UNION ALL
+
+        SELECT * FROM {base_table}
+        WHERE auto_id IN ({', '.join(map(str, replicate_ids))})
+    """
+
+    training_view = TrainingView.create_with_sql(fs, sql_query=custom_sql, id_column="auto_id")
+    df = training_view.pull_dataframe()
+    print(f"Shape with custom SQL: {df.shape}")
+    print(df["training"].value_counts())
+
+    # Verify oversampling - check if replicated IDs appear twice
+    for rep_id in replicate_ids:
+        count = len(df[df["auto_id"] == rep_id])
+        print(f"ID {rep_id} appears {count} times")

workbench/core/views/view.py

@@ -91,11 +91,11 @@ class View:
             self.table, self.data_source.database, self.data_source.boto3_session
         )
 
-    def pull_dataframe(self, limit: int = 50000) -> Union[pd.DataFrame, None]:
+    def pull_dataframe(self, limit: int = 100000) -> Union[pd.DataFrame, None]:
         """Pull a DataFrame based on the view type
 
         Args:
-            limit (int): The maximum number of rows to pull (default: 50000)
+            limit (int): The maximum number of rows to pull (default: 100000)
 
         Returns:
             Union[pd.DataFrame, None]: The DataFrame for the view or None if it doesn't exist
@@ -196,12 +196,52 @@ class View:
 
         # The BaseView always exists
         if self.view_name == "base":
-            return True
+            return
 
         # Check the database directly
         if not self._check_database():
             self._auto_create_view()
 
+    def copy(self, dest_view_name: str) -> "View":
+        """Copy this view to a new view with a different name
+
+        Args:
+            dest_view_name (str): The destination view name (e.g. "training_v1")
+
+        Returns:
+            View: A new View object for the destination view
+        """
+        # Can't copy the base view
+        if self.view_name == "base":
+            self.log.error("Cannot copy the base view")
+            return None
+
+        # Get the view definition
+        get_view_query = f"""
+        SELECT view_definition
+        FROM information_schema.views
+        WHERE table_schema = '{self.database}'
+        AND table_name = '{self.table}'
+        """
+        df = self.data_source.query(get_view_query)
+
+        if df.empty:
+            self.log.error(f"View {self.table} not found")
+            return None
+
+        view_definition = df.iloc[0]["view_definition"]
+
+        # Create the new view with the destination name
+        dest_table = f"{self.base_table_name}___{dest_view_name.lower()}"
+        create_view_query = f'CREATE OR REPLACE VIEW "{dest_table}" AS {view_definition}'
+
+        self.log.important(f"Copying view {self.table} to {dest_table}...")
+        self.data_source.execute_statement(create_view_query)
+
+        # Return a new View object for the destination
+        artifact = FeatureSet(self.artifact_name) if self.is_feature_set else DataSource(self.artifact_name)
+        return View(artifact, dest_view_name, auto_create_view=False)
+
     def _check_database(self) -> bool:
         """Internal: Check if the view exists in the database
 
@@ -324,3 +364,13 @@ if __name__ == "__main__":
     # Test supplemental data tables deletion
     view = View(fs, "test_view")
     view.delete()
+
+    # Test copying a view
+    fs = FeatureSet("test_features")
+    display_view = View(fs, "display")
+    copied_view = display_view.copy("display_copy")
+    print(copied_view)
+    print(copied_view.pull_dataframe().head())
+
+    # Clean up copied view
+    copied_view.delete()