workbench 0.8.158__py3-none-any.whl → 0.8.159__py3-none-any.whl

workbench/api/feature_set.py

@@ -87,8 +87,9 @@ class FeatureSet(FeatureSetCore):
         model_import_str: str = None,
         custom_script: Union[str, Path] = None,
         custom_args: dict = None,
+        training_image: str = "xgb_training",
+        inference_image: str = "xgb_inference",
         inference_arch: str = "x86_64",
-        inference_image: str = "inference",
         **kwargs,
     ) -> Union[Model, None]:
         """Create a Model from the FeatureSet
@@ -101,11 +102,12 @@ class FeatureSet(FeatureSetCore):
             description (str, optional): Set the description for the model. If not give a description is generated.
             feature_list (list, optional): Set the feature list for the model. If not given a feature list is generated.
             target_column (str, optional): The target column for the model (use None for unsupervised model)
-            model_class (str, optional): Scikit model class to use (e.g. "KMeans", default: None)
+            model_class (str, optional): Model class to use (e.g. "KMeans", "PyTorch", default: None)
             model_import_str (str, optional): The import for the model (e.g. "from sklearn.cluster import KMeans")
             custom_script (str, optional): The custom script to use for the model (default: None)
+            training_image (str, optional): The training image to use (default: "xgb_training")
+            inference_image (str, optional): The inference image to use (default: "xgb_inference")
             inference_arch (str, optional): The architecture to use for inference (default: "x86_64")
-            inference_image (str, optional): The inference image to use (default: "inference")
 
         Returns:
             Model: The Model created from the FeatureSet (or None if the Model could not be created)
@@ -125,6 +127,11 @@ class FeatureSet(FeatureSetCore):
         # Create the Model Tags
         tags = [name] if tags is None else tags
 
+        # If the model_class is PyTorch, ensure we set the training and inference images
+        if model_class and model_class.lower() == "pytorch":
+            training_image = "pytorch_training"
+            inference_image = "pytorch_inference"
+
         # Transform the FeatureSet into a Model
         features_to_model = FeaturesToModel(
             feature_name=self.name,
@@ -134,8 +141,9 @@ class FeatureSet(FeatureSetCore):
             model_import_str=model_import_str,
             custom_script=custom_script,
             custom_args=custom_args,
-            inference_arch=inference_arch,
+            training_image=training_image,
             inference_image=inference_image,
+            inference_arch=inference_arch,
         )
         features_to_model.set_output_tags(tags)
         features_to_model.transform(

workbench/api/meta.py

@@ -28,7 +28,7 @@ class Meta(CloudMeta):
        meta.data_sources()
        meta.feature_sets(details=True/False)
        meta.models(details=True/False)
-       meta.endpoints()
+       meta.endpoints(details=True/False)
        meta.views()
        meta.pipelines()
 

workbench/cached/cached_feature_set.py

@@ -79,6 +79,7 @@ if __name__ == "__main__":
 
     # Retrieve an existing FeatureSet
     my_features = CachedFeatureSet("abalone_features")
+    pprint(my_features.smart_sample())
     pprint(my_features.summary())
     pprint(my_features.details())
     pprint(my_features.health_check())

workbench/cached/cached_meta.py

@@ -13,8 +13,6 @@ from workbench.utils.workbench_cache import WorkbenchCache
 
 
 # Decorator to cache method results from the Meta class
-# Note: This has to be outside the class definition to work properly in Python 3.9
-#       When we deprecated support for 3.9, move this back into the class definition
 def cache_result(method):
     """Decorator to cache method results in meta_cache"""
 
@@ -24,11 +22,8 @@ def cache_result(method):
         cache_key = CachedMeta._flatten_redis_key(method, *args, **kwargs)
 
         # Check for fresh data, spawn thread to refresh if stale
-        if WorkbenchCache.refresh_enabled and self.fresh_cache.get(cache_key) is None:
-            self.log.debug(f"Async: Metadata for {cache_key} refresh thread started...")
-            self.fresh_cache.set(cache_key, True)  # Mark as refreshed
-
-            # Spawn a thread to refresh data without blocking
+        if self.fresh_cache.atomic_set(cache_key, True):
+            self.log.important(f"Async: Metadata for {cache_key} refresh thread started...")
             self.thread_pool.submit(self._refresh_data_in_background, cache_key, method, *args, **kwargs)
 
         # Return data (fresh or stale) if available
@@ -62,7 +57,7 @@ class CachedMeta(CloudMeta):
        meta.data_sources()
        meta.feature_sets(details=True/False)
        meta.models(details=True/False)
-       meta.endpoints()
+       meta.endpoints(details=True/False)
        meta.views()
 
        # These are 'describe' methods
@@ -91,7 +86,7 @@ class CachedMeta(CloudMeta):
 
         # Create both our Meta Cache and Fresh Cache (tracks if data is stale)
         self.meta_cache = WorkbenchCache(prefix="meta")
-        self.fresh_cache = WorkbenchCache(prefix="meta_fresh", expire=90)  # 90-second expiration
+        self.fresh_cache = WorkbenchCache(prefix="meta_fresh", expire=300)  # 5-minute expiration
 
         # Create a ThreadPoolExecutor for refreshing stale data
         self.thread_pool = ThreadPoolExecutor(max_workers=5)
@@ -193,13 +188,16 @@ class CachedMeta(CloudMeta):
         return super().models(details=details)
 
     @cache_result
-    def endpoints(self) -> pd.DataFrame:
+    def endpoints(self, details: bool = False) -> pd.DataFrame:
         """Get a summary of the Endpoints deployed in the Cloud Platform
 
+        Args:
+            details (bool, optional): Include detailed information. Defaults to False.
+
         Returns:
             pd.DataFrame: A summary of the Endpoints in the Cloud Platform
         """
-        return super().endpoints()
+        return super().endpoints(details=details)
 
     @cache_result
     def glue_job(self, job_name: str) -> Union[dict, None]:
@@ -266,7 +264,7 @@ class CachedMeta(CloudMeta):
         """Background task to refresh AWS metadata."""
         result = method(self, *args, **kwargs)
         self.meta_cache.set(cache_key, result)
-        self.log.debug(f"Updated Metadata for {cache_key}")
+        self.log.important(f"Updated Metadata for {cache_key}")
 
     @staticmethod
     def _flatten_redis_key(method, *args, **kwargs):

workbench/core/artifacts/cached_artifact_mixin.py

@@ -13,7 +13,7 @@ class CachedArtifactMixin:
     # Class-level caches, thread pool, and shutdown flag
     log = logging.getLogger("workbench")
     artifact_cache = WorkbenchCache(prefix="artifact_cache")
-    fresh_cache = WorkbenchCache(prefix="artifact_fresh_cache", expire=10)
+    fresh_cache = WorkbenchCache(prefix="artifact_fresh_cache", expire=120)
     thread_pool = ThreadPoolExecutor(max_workers=5)
 
     @staticmethod
@@ -45,8 +45,8 @@ class CachedArtifactMixin:
                 cls.artifact_cache.set(cache_key, result)
                 return result
 
-            # Stale cache: Refresh in the background if enabled and no refresh is already in progress
-            if WorkbenchCache.refresh_enabled and cache_fresh is None:
+            # Stale cache: Refresh in the background
+            if cache_fresh is None:
                 self.log.debug(f"Async: Refresh thread started: {cache_key}...")
                 cls.fresh_cache.set(cache_key, True)
                 cls.thread_pool.submit(cls._refresh_data_in_background, self, cache_key, method, *args, **kwargs)
@@ -88,4 +88,7 @@ if __name__ == "__main__":
     my_model = CachedModel("abalone-regression")
     pprint(my_model.summary())
     pprint(my_model.details())
+    # Second call to demonstrate caching
+    pprint(my_model.summary())
+    pprint(my_model.details())
     CachedArtifactMixin._shutdown()

workbench/core/artifacts/model_core.py

@@ -42,24 +42,36 @@ class ModelImages:
 
     image_uris = {
         # US East 1 images
-        ("us-east-1", "training", "0.1", "x86_64"): (
+        ("us-east-1", "xgb_training", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
         ),
-        ("us-east-1", "inference", "0.1", "x86_64"): (
+        ("us-east-1", "xgb_inference", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
         ),
+        ("us-east-1", "pytorch_training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-pytorch-training:0.1"
+        ),
+        ("us-east-1", "pytorch_inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-pytorch-inference:0.1"
+        ),
         # US West 2 images
-        ("us-west-2", "training", "0.1", "x86_64"): (
+        ("us-west-2", "xgb_training", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
         ),
-        ("us-west-2", "inference", "0.1", "x86_64"): (
+        ("us-west-2", "xgb_inference", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
         ),
+        ("us-west-2", "pytorch_training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-pytorch-training:0.1"
+        ),
+        ("us-west-2", "pytorch_inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-pytorch-inference:0.1"
+        ),
         # ARM64 images
-        ("us-east-1", "inference", "0.1", "arm64"): (
+        ("us-east-1", "xgb_inference", "0.1", "arm64"): (
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
         ),
-        ("us-west-2", "inference", "0.1", "arm64"): (
+        ("us-west-2", "xgb_inference", "0.1", "arm64"): (
             "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
         ),
         # Meta Endpoint inference images
@@ -72,7 +84,7 @@ class ModelImages:
     }
 
     @classmethod
-    def get_image_uri(cls, region, image_type="training", version="0.1", architecture="x86_64"):
+    def get_image_uri(cls, region, image_type, version="0.1", architecture="x86_64"):
         key = (region, image_type, version, architecture)
         if key in cls.image_uris:
             return cls.image_uris[key]

workbench/core/cloud_platform/aws/aws_meta.py

@@ -179,7 +179,7 @@ class AWSMeta:
                     feature_set_details.update(self.sm_client.describe_feature_group(FeatureGroupName=name))
 
                 # Retrieve Workbench metadata from tags
-                aws_tags = self.get_aws_tags(fg["FeatureGroupArn"])
+                aws_tags = self.get_aws_tags(fg["FeatureGroupArn"]) if details else {}
                 summary = {
                     "Feature Group": name,
                     "Health": "",
@@ -258,70 +258,60 @@ class AWSMeta:
         df = pd.DataFrame(model_summary).convert_dtypes()
         return df.sort_values(by="Created", ascending=False)
 
-    def endpoints(self, refresh: bool = False) -> pd.DataFrame:
+    def endpoints(self, details: bool = False) -> pd.DataFrame:
         """Get a summary of the Endpoints in AWS.
 
         Args:
-            refresh (bool, optional): Force a refresh of the metadata. Defaults to False.
+            details (bool, optional): Get additional details (Defaults to False).
 
         Returns:
             pd.DataFrame: A summary of the Endpoints in AWS.
         """
         from workbench.utils.endpoint_utils import is_monitored  # noqa: E402
 
-        # Initialize the SageMaker client and list all endpoints
-        sagemaker_client = self.boto3_session.client("sagemaker")
-        paginator = sagemaker_client.get_paginator("list_endpoints")
+        # Use our SageMaker client to list all endpoints
+        paginator = self.sm_client.get_paginator("list_endpoints")
         data_summary = []
 
         # Use the paginator to retrieve all endpoints
         for page in paginator.paginate():
             for endpoint in page["Endpoints"]:
                 endpoint_name = endpoint["EndpointName"]
-                endpoint_info = sagemaker_client.describe_endpoint(EndpointName=endpoint_name)
 
-                # Retrieve Workbench metadata from tags
-                aws_tags = self.get_aws_tags(endpoint_info["EndpointArn"])
-                health_tags = aws_tags.get("workbench_health_tags", "")
-
-                # Retrieve endpoint configuration to determine instance type or serverless info
-                endpoint_config_name = endpoint_info["EndpointConfigName"]
-
-                # Getting the endpoint configuration can fail so account for that
-                try:
-                    endpoint_config = sagemaker_client.describe_endpoint_config(EndpointConfigName=endpoint_config_name)
-                    production_variant = endpoint_config["ProductionVariants"][0]
-                    # Determine instance type or serverless configuration
-                    instance_type = production_variant.get("InstanceType")
-                    if instance_type is None:
-                        # If no instance type, it's a serverless configuration
-                        mem_size = production_variant["ServerlessConfig"]["MemorySizeInMB"]
-                        concurrency = production_variant["ServerlessConfig"]["MaxConcurrency"]
-                        instance_type = f"Serverless ({mem_size // 1024}GB/{concurrency})"
-                except sagemaker_client.exceptions.ClientError:
-                    # If the endpoint config is not found, change the config name to reflect this
-                    endpoint_config_name = f"{endpoint_config_name} (Not Found)"
-                    production_variant = {}
-                    instance_type = "Unknown"
-
-                # Check if the endpoint has monitoring enabled
-                endpoint_monitored = is_monitored(endpoint_name, sagemaker_client)
+                # Grab various endpoint details
+                endpoint_details = {"config": {"instance": "-", "variant": "-"}, "monitored": "-"}
+                aws_tags = {}
+                if details:
+                    endpoint_details = self.sm_client.describe_endpoint(EndpointName=endpoint_name)
+
+                    # Retrieve AWS Tags for this Endpoint
+                    aws_tags = self.get_aws_tags(endpoint_details["EndpointArn"])
+
+                    # Getting the endpoint configuration
+                    config_info = self._endpoint_config_info(endpoint_details["EndpointConfigName"])
+                    endpoint_details["config"] = config_info
+
+                    # Check if the endpoint has monitoring enabled
+                    endpoint_details["monitored"] = is_monitored(endpoint_name, self.sm_client)
 
                 # Compile endpoint summary
+                created = (
+                    datetime_string(endpoint_details["CreationTime"]) if "CreationTime" in endpoint_details else "-"
+                )
                 summary = {
                     "Name": endpoint_name,
-                    "Health": health_tags,
+                    "Health": aws_tags.get("workbench_health_tags", ""),
                     "Owner": aws_tags.get("workbench_owner", "-"),
-                    "Instance": instance_type,
-                    "Created": datetime_string(endpoint_info.get("CreationTime")),
+                    "Instance": endpoint_details["config"]["instance"],
+                    "Created": created,
                     "Input": aws_tags.get("workbench_input", "-"),
-                    "Status": endpoint_info["EndpointStatus"],
-                    "Config": endpoint_config_name,
-                    "Variant": production_variant.get("VariantName", "-"),
-                    "Capture": str(endpoint_info.get("DataCaptureConfig", {}).get("EnableCapture", "False")),
-                    "Samp(%)": str(endpoint_info.get("DataCaptureConfig", {}).get("CurrentSamplingPercentage", "-")),
+                    "Status": endpoint_details.get("EndpointStatus", "-"),
+                    "Config": endpoint_details.get("EndpointConfigName", "-"),
+                    "Variant": endpoint_details["config"]["variant"],
+                    "Capture": str(endpoint_details.get("DataCaptureConfig", {}).get("EnableCapture", "False")),
+                    "Samp(%)": str(endpoint_details.get("DataCaptureConfig", {}).get("CurrentSamplingPercentage", "-")),
                     "Tags": aws_tags.get("workbench_tags", "-"),
-                    "Monitored": endpoint_monitored,
+                    "Monitored": endpoint_details["monitored"],
                 }
                 data_summary.append(summary)
 
@@ -329,6 +319,34 @@ class AWSMeta:
         df = pd.DataFrame(data_summary).convert_dtypes()
         return df.sort_values(by="Created", ascending=False)
 
+    def _endpoint_config_info(self, endpoint_config_name: str) -> dict:
+        """Internal: Get the Endpoint Configuration information for the given endpoint config name.
+
+        Args:
+            endpoint_config_name (str): The name of the endpoint configuration.
+
+        Returns:
+            dict: The endpoint configuration details.
+        """
+
+        # Retrieve the endpoint configuration
+        try:
+            endpoint_config = self.sm_client.describe_endpoint_config(EndpointConfigName=endpoint_config_name)
+            production_variant = endpoint_config["ProductionVariants"][0]
+
+            # Determine instance type or serverless configuration
+            instance_type = production_variant.get("InstanceType")
+            if instance_type is None:
+                # If no instance type, it's a serverless configuration
+                mem_size = production_variant["ServerlessConfig"]["MemorySizeInMB"]
+                concurrency = production_variant["ServerlessConfig"]["MaxConcurrency"]
+                instance_type = f"Serverless ({mem_size // 1024}GB/{concurrency})"
+
+            return {"instance": instance_type, "variant": production_variant.get("VariantName", "-")}
+        except self.sm_client.exceptions.ClientError as e:
+            self.log.error(f"Error retrieving endpoint config {endpoint_config_name}: {e}")
+            return {"instance": "-", "variant": "-"}
+
     def pipelines(self) -> pd.DataFrame:
         """List all the Pipelines in the S3 Bucket
 
@@ -702,7 +720,6 @@ class AWSMeta:
 
 if __name__ == "__main__":
     """Exercise the Workbench AWSMeta Class"""
-    import time
     from pprint import pprint
 
     # Pandas Display Options
@@ -712,6 +729,7 @@ if __name__ == "__main__":
     # Create the class
     meta = AWSMeta()
 
+    """
     # Test the __repr__ method
     print(meta)
 
@@ -759,11 +777,15 @@ if __name__ == "__main__":
     start_time = time.time()
     pprint(meta.models(details=True))
     print(f"Elapsed Time Model (with details): {time.time() - start_time:.2f}")
-
+    """
     # Get the Endpoints
     print("\n\n*** Endpoints ***")
     pprint(meta.endpoints())
 
+    # Get the Endpoints with Details
+    print("\n\n*** Endpoints with Details ***")
+    pprint(meta.endpoints(details=True))
+
     # List Pipelines
     print("\n\n*** Workbench Pipelines ***")
     pprint(meta.pipelines())
@@ -785,7 +807,6 @@ if __name__ == "__main__":
     pprint(meta.model("abalone-regression"))
     print("\n\n*** Endpoint Details ***")
     pprint(meta.endpoint("abalone-regression"))
-    pprint(meta.endpoint("test-timing-realtime"))
 
     # Test out a non-existent model
     print("\n\n*** Model Doesn't Exist ***")

workbench/core/cloud_platform/cloud_meta.py

@@ -121,13 +121,16 @@ class CloudMeta(AWSMeta):
         """
         return super().models(details=details)
 
-    def endpoints(self) -> pd.DataFrame:
+    def endpoints(self, details: bool = False) -> pd.DataFrame:
         """Get a summary of the Endpoints deployed in the Cloud Platform
 
+        Args:
+            details (bool, optional): Include detailed information. Defaults to False.
+
         Returns:
             pd.DataFrame: A summary of the Endpoints in the Cloud Platform
         """
-        return super().endpoints()
+        return super().endpoints(details=details)
 
     def pipelines(self) -> pd.DataFrame:
         """Get a summary of the Pipelines deployed in the Cloud Platform

workbench/core/transforms/features_to_model/features_to_model.py

@@ -37,8 +37,9 @@ class FeaturesToModel(Transform):
         model_import_str=None,
         custom_script=None,
         custom_args=None,
+        training_image="xgb_training",
+        inference_image="xgb_inference",
         inference_arch="x86_64",
-        inference_image="inference",
     ):
         """FeaturesToModel Initialization
         Args:
@@ -49,8 +50,9 @@ class FeaturesToModel(Transform):
             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")
             inference_arch (str, optional): Inference architecture (default "x86_64")
-            inference_image (str, optional): Inference image (default "inference")
         """
 
         # Make sure the model_name is a valid name
@@ -73,8 +75,9 @@ class FeaturesToModel(Transform):
         self.model_feature_list = None
         self.target_column = None
         self.class_labels = None
-        self.inference_arch = inference_arch
+        self.training_image = training_image
         self.inference_image = inference_image
+        self.inference_arch = inference_arch
 
     def transform_impl(
         self, target_column: str, description: str = None, feature_list: list = None, train_all_data=False, **kwargs
@@ -229,7 +232,7 @@ 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, "training", "0.1")
+        image = ModelImages.get_image_uri(self.sm_session.boto_region_name, self.training_image, "0.1")
         self.estimator = Estimator(
             entry_point=entry_point,
             source_dir=source_dir,
@@ -246,6 +249,7 @@ class FeaturesToModel(Transform):
         training_job_name = f"{self.output_name}-{training_date_time_utc}"
 
         # Train the estimator
+        self.log.important(f"Training the Model {self.output_name} with Training Image {image}...")
         self.estimator.fit({"train": s3_training_path}, job_name=training_job_name)
 
         # Now delete the training data
@@ -297,7 +301,7 @@ class FeaturesToModel(Transform):
         image = ModelImages.get_image_uri(
             self.sm_session.boto_region_name, self.inference_image, "0.1", self.inference_arch
         )
-        self.log.important(f"Registering model {self.output_name} with image {image}...")
+        self.log.important(f"Registering model {self.output_name} with Inference Image {image}...")
         model = self.estimator.create_model(role=self.workbench_role_arn)
         if aws_region:
             self.log.important(f"Setting AWS Region: {aws_region} for model {self.output_name}...")

workbench/core/transforms/model_to_endpoint/model_to_endpoint.py

@@ -78,6 +78,12 @@ class ModelToEndpoint(Transform):
             sagemaker_session=self.sm_session,
         )
 
+        # Log the image that will be used for deployment
+        inference_image = self.sm_client.describe_model_package(ModelPackageName=model_package_arn)[
+            "InferenceSpecification"
+        ]["Containers"][0]["Image"]
+        self.log.important(f"Deploying Model Package: {self.input_name} with Inference Image: {inference_image}")
+
         # Get the metadata/tags to push into AWS
         aws_tags = self.get_aws_tags()