Merge branch 'master' into fix/remove-unnecessary-get-caller-identity-call

Author: evakravi

CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## v2.169.0 (2023-06-29)
+
+### Features
+
+ * Add support for tags in to_pipeline API for feature processor
+ * model registry integration to model cards to support model packages
+ * SDK Defaults - DebugHookConfig defaults in TrainingJob API
+ * Add segment config for Clarify
+
+### Bug Fixes and Other Changes
+
+ * Neuronx image retrieval missing sdk information
+
+### Documentation Changes
+
+ * Doc updates for SDK defaults - S3 Params, Env Variables, Disable Profiler, and DebugHookConfig
+
 ## v2.168.0 (2023-06-22)
 
 ### Features

VERSION

@@ -1 +1 @@
-2.168.1.dev0
+2.169.1.dev0

src/sagemaker/feature_store/feature_processor/_constants.py

@@ -40,3 +40,8 @@
 S3_DATA_DISTRIBUTION_TYPE = "FullyReplicated"
 PIPELINE_CONTEXT_NAME_TAG_KEY = "sm-fs-fe:feature-engineering-pipeline-context-name"
 PIPELINE_VERSION_CONTEXT_NAME_TAG_KEY = "sm-fs-fe:feature-engineering-pipeline-version-context-name"
+TO_PIPELINE_RESERVED_TAG_KEYS = [
+    FEATURE_PROCESSOR_TAG_KEY,
+    PIPELINE_CONTEXT_NAME_TAG_KEY,
+    PIPELINE_VERSION_CONTEXT_NAME_TAG_KEY,
+]

src/sagemaker/feature_store/feature_processor/feature_scheduler.py

@@ -16,7 +16,7 @@
 import json
 import re
 from datetime import datetime
-from typing import Callable, List, Optional, Dict, Sequence, Union, Any
+from typing import Callable, List, Optional, Dict, Sequence, Union, Any, Tuple
 
 import pytz
 from botocore.exceptions import ClientError
@@ -58,6 +58,7 @@
     PIPELINE_NAME_MAXIMUM_LENGTH,
     RESOURCE_NOT_FOUND,
     FEATURE_GROUP_ARN_REGEX_PATTERN,
+    TO_PIPELINE_RESERVED_TAG_KEYS,
 )
 from sagemaker.feature_store.feature_processor._feature_processor_config import (
     FeatureProcessorConfig,
@@ -107,6 +108,7 @@ def to_pipeline(
     role: Optional[str] = None,
     transformation_code: Optional[TransformationCode] = None,
     max_retries: Optional[int] = None,
+    tags: Optional[List[Tuple[str, str]]] = None,
     sagemaker_session: Optional[Session] = None,
 ) -> str:
     """Creates a sagemaker pipeline that takes in a callable as a training step.
@@ -127,6 +129,8 @@ def to_pipeline(
             code for Lineage tracking. This code is not used for actual transformation.
         max_retries (Optional[int]): The number of times to retry sagemaker pipeline step.
             If not specified, sagemaker pipline step will not retry.
+        tags (List[Tuple[str, str]): A list of tags attached to the pipeline. If not specified,
+            no custom tags will be attached to the pipeline.
         sagemaker_session (Optional[Session]): Session object which manages interactions
             with Amazon SageMaker APIs and any other AWS services needed. If not specified, the
             function creates one using the default AWS configuration chain.
@@ -135,6 +139,8 @@ def to_pipeline(
     """
 
     _validate_input_for_to_pipeline_api(pipeline_name, step)
+    if tags:
+        _validate_tags_for_to_pipeline_api(tags)
 
     _sagemaker_session = sagemaker_session or Session()
 
@@ -200,12 +206,15 @@ def to_pipeline(
         sagemaker_session=_sagemaker_session,
         parameters=[SCHEDULED_TIME_PIPELINE_PARAMETER],
     )
+    pipeline_tags = [dict(Key=FEATURE_PROCESSOR_TAG_KEY, Value=FEATURE_PROCESSOR_TAG_VALUE)]
+    if tags:
+        pipeline_tags.extend([dict(Key=k, Value=v) for k, v in tags])
 
     pipeline = Pipeline(**pipeline_request_dict)
     logger.info("Creating/Updating sagemaker pipeline %s", pipeline_name)
     pipeline.upsert(
         role_arn=_role,
-        tags=[dict(Key=FEATURE_PROCESSOR_TAG_KEY, Value=FEATURE_PROCESSOR_TAG_VALUE)],
+        tags=pipeline_tags,
     )
     logger.info("Created sagemaker pipeline %s", pipeline_name)
 
@@ -514,6 +523,23 @@ def _validate_input_for_to_pipeline_api(pipeline_name: str, step: Callable) -> N
         )
 
 
+def _validate_tags_for_to_pipeline_api(tags: List[Tuple[str, str]]) -> None:
+    """Validate tags provided to to_pipeline API.
+
+    Args:
+        tags (List[Tuple[str, str]]): A list of tags attached to the pipeline.
+
+    Raises (ValueError): raises ValueError when any of the following scenario happen:
+           1. reserved tag keys are provided to API.
+    """
+    provided_tag_keys = [tag_key_value_pair[0] for tag_key_value_pair in tags]
+    for reserved_tag_key in TO_PIPELINE_RESERVED_TAG_KEYS:
+        if reserved_tag_key in provided_tag_keys:
+            raise ValueError(
+                f"{reserved_tag_key} is a reserved tag key for to_pipeline API. Please choose another tag."
+            )
+
+
 def _validate_lineage_resources_for_to_pipeline_api(
     feature_processor_config: FeatureProcessorConfig, sagemaker_session: Session
 ) -> None:

src/sagemaker/model_card/__init__.py

@@ -28,6 +28,7 @@
     EvaluationJob,
     AdditionalInformation,
     ModelCard,
+    ModelPackage,
 )
 
 from sagemaker.model_card.schema_constraints import (  # noqa: F401 # pylint: disable=unused-import

src/sagemaker/model_card/evaluation_metric_parsers.py

@@ -31,6 +31,7 @@ class EvaluationMetricTypeEnum(str, Enum):
     MODEL_CARD_METRIC_SCHEMA = "Model Card Metric Schema"
     CLARIFY_BIAS = "Clarify Bias"
     CLARIFY_EXPLAINABILITY = "Clarify Explainability"
+    MODEL_MONITOR_MODEL_QUALITY = "Model Monitor Model Quality"
     REGRESSION = "Model Monitor Model Quality Regression"
     BINARY_CLASSIFICATION = "Model Monitor Model Quality Binary Classification"
     MULTICLASS_CLASSIFICATION = "Model Monitor Model Quality Multiclass Classification"
@@ -138,6 +139,7 @@ def _parse(self, json_data: dict):
                             [
                                 {"name": i["name"], "value": i["value"], "type": "number"}
                                 for i in item["metrics"]
+                                if i["value"] is not None
                             ]
                         )
                     for group_name, metric_data in group_data.items():
@@ -368,9 +370,10 @@ def _parse(self, json_data: dict):
         result = {"metric_groups": []}
         for group_name, group_data in json_data.items():
             metric_data = []
-            for metric_name, raw_data in group_data.item():
-                metric_data.extend(self._parse_basic_metric(metric_name, raw_data))
-            result["metric_groups"].append({"name": group_name, "metric_data": metric_data})
+            if group_name == "regression_metrics":
+                for metric_name, raw_data in group_data.items():
+                    metric_data.extend(self._parse_basic_metric(metric_name, raw_data))
+                result["metric_groups"].append({"name": group_name, "metric_data": metric_data})
         return result
 
 
@@ -388,7 +391,7 @@ def _validate(self, json_data: dict):
         """
         if (
             "binary_classification_metrics" not in json_data
-            and "multiclass_classification_metrics" in json_data
+            and "multiclass_classification_metrics" not in json_data
         ):
             raise ValueError("Missing *_classification_metrics from the metric data.")
 
@@ -401,6 +404,11 @@ def _parse(self, json_data: dict):
         result = {"metric_groups": []}
         for group_name, group_data in json_data.items():
             metric_data = []
+            if group_name not in (
+                "binary_classification_metrics",
+                "multiclass_classification_metrics",
+            ):
+                continue
             for metric_name, raw_data in group_data.items():
                 metric_data.extend(self._parse_confusion_matrix(metric_name, raw_data))
                 metric_data.extend(
@@ -506,11 +514,45 @@ def _parse_precision_recall_curve(self, metric_name, raw_data):
         return metric_data
 
 
+class ModelMonitorModelQualityParser(ParserBase):
+    """Top level parser for model monitor model quality metric type"""
+
+    def _validate(self, json_data: dict):
+        """Implement ParserBase._validate.
+
+        Args:
+            json_data (dict): Metric data to be validated.
+
+        Raises:
+            ValueError: missing model monitor model quality metrics.
+        """
+        if len(json_data) == 0:
+            raise ValueError("Missing model monitor model quality metrics from the metric data.")
+
+    def _parse(self, json_data: dict):
+        """Implement ParserBase._parse.
+
+        Args:
+            json_data (dict): Raw metric data.
+        """
+        result = {"metric_groups": []}
+        if "regression_metrics" in json_data:
+            result = RegressionParser().run(json_data)
+        elif (
+            "binary_classification_metrics" in json_data
+            or "multiclass_classification_metrics" in json_data
+        ):
+            result = ClassificationParser().run(json_data)
+
+        return result
+
+
 EVALUATION_METRIC_PARSERS = {
     EvaluationMetricTypeEnum.MODEL_CARD_METRIC_SCHEMA: DefaultParser(),
     EvaluationMetricTypeEnum.CLARIFY_BIAS: ClarifyBiasParser(),
     EvaluationMetricTypeEnum.CLARIFY_EXPLAINABILITY: ClarifyExplainabilityParser(),
     EvaluationMetricTypeEnum.REGRESSION: RegressionParser(),
     EvaluationMetricTypeEnum.BINARY_CLASSIFICATION: ClassificationParser(),
     EvaluationMetricTypeEnum.MULTICLASS_CLASSIFICATION: ClassificationParser(),
+    EvaluationMetricTypeEnum.MODEL_MONITOR_MODEL_QUALITY: ModelMonitorModelQualityParser(),
 }

src/sagemaker/model_card/helpers.py

@@ -62,13 +62,21 @@ def _clean_descriptor_name(self, name: str):
 
         return name
 
+    def _skip_encoding(self, attr: str):
+        """Skip encoding if the attribute is an instance of _SkipEncodingDecoding descriptor"""
+        if attr in self.__class__.__dict__:
+            return isinstance(self.__class__.__dict__[attr], _SkipEncodingDecoding)
+
+        return False
+
     def _to_request_dict(self):
         """Implement this method in a subclass to return a custom request_dict."""
         request_data = {}
         for attr, value in self.__dict__.items():
             if value is not None:
                 name = self._clean_descriptor_name(attr)
-                request_data[name] = value
+                if not self._skip_encoding(name):
+                    request_data[name] = value
 
         return request_data
 
@@ -149,6 +157,38 @@ def decode(self, value: dict):
         pass  # pylint: disable=W0107
 
 
+class _SkipEncodingDecoding(_DescriptorBase):
+    """Object that skip the encoding/decoding in model card attributes."""
+
+    def __init__(self, value_type: Any):
+        """Initialize an SkipEncodingDecoding descriptor.
+
+        Args:
+            value_type (Any): Value type of the attribute.
+        """
+        self.value_type = value_type
+
+    def validate(self, value: Any):
+        """Check if value type is valid.
+
+        Args:
+            value (Any): value type depends on self.value_type
+
+        Raises:
+            ValueError: value is not a self.value_type.
+        """
+        if value is not None and not isinstance(value, self.value_type):
+            raise ValueError(f"Please assign a {self.value_type} to {self.private_name[1:]}")
+
+    def require_decode(self, value: Any):
+        """No decoding is required."""
+        return False
+
+    def decode(self, value: Any):
+        """No decoding is required. Required placeholder for abstractmethod"""
+        pass  # pylint: disable=W0107
+
+
 class _OneOf(_DescriptorBase):
     """Verifies that a value is one of a restricted set of options"""
 
@@ -463,9 +503,12 @@ def _read_s3_json(session: Session, bucket: str, key: str):
             raise
 
     result = {}
-    if data["ContentType"] == "application/json":
+    if data["ContentType"] == "application/json" or data["ContentType"] == "binary/octet-stream":
         result = json.loads(data["Body"].read().decode("utf-8"))
     else:
-        logger.warning("Invalid file type %s. application/json is expected.", data["ContentType"])
+        logger.warning(
+            "Invalid file type %s. application/json or binary/octet-stream is expected.",
+            data["ContentType"],
+        )
 
     return result