Merge branch 'master' into fix-pydoc

Author: shreyapandit

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## v2.49.2 (2021-07-21)
+
+### Bug Fixes and Other Changes
+
+ * order of populating container list
+ * upgrade Adobe Analytics cookie to 3.0
+
 ## v2.49.1 (2021-07-19)
 
 ### Bug Fixes and Other Changes

VERSION

@@ -1 +1 @@
-2.49.2.dev0
+2.49.3.dev0

src/sagemaker/model.py

@@ -13,6 +13,7 @@
 """Placeholder docstring"""
 from __future__ import absolute_import
 
+import abc
 import json
 import logging
 import os
@@ -29,6 +30,7 @@
     git_utils,
 )
 from sagemaker.deprecations import removed_kwargs
+from sagemaker.predictor import PredictorBase
 from sagemaker.transformer import Transformer
 
 LOGGER = logging.getLogger("sagemaker")
@@ -38,7 +40,23 @@
 )
 
 
-class Model(object):
+class ModelBase(abc.ABC):
+    """An object that encapsulates a trained model.
+
+    Models can be deployed to compute services like a SageMaker ``Endpoint``
+    or Lambda. Deployed models can be used to perform real-time inference.
+    """
+
+    @abc.abstractmethod
+    def deploy(self, *args, **kwargs) -> PredictorBase:
+        """Deploy this model to a compute service."""
+
+    @abc.abstractmethod
+    def delete_model(self, *args, **kwargs) -> None:
+        """Destroy resources associated with this model."""
+
+
+class Model(ModelBase):
     """A SageMaker ``Model`` that can be deployed to an ``Endpoint``."""
 
     def __init__(

src/sagemaker/predictor.py

@@ -13,6 +13,9 @@
 """Placeholder docstring"""
 from __future__ import print_function, absolute_import
 
+import abc
+from typing import Any, Tuple
+
 from sagemaker.deprecations import (
     deprecated_class,
     deprecated_deserialize,
@@ -51,7 +54,29 @@
 from sagemaker.lineage.context import EndpointContext
 
 
-class Predictor(object):
+class PredictorBase(abc.ABC):
+    """An object that encapsulates a deployed model."""
+
+    @abc.abstractmethod
+    def predict(self, *args, **kwargs) -> Any:
+        """Perform inference on the provided data and return a prediction."""
+
+    @abc.abstractmethod
+    def delete_endpoint(self, *args, **kwargs) -> None:
+        """Destroy resources associated with this predictor."""
+
+    @property
+    @abc.abstractmethod
+    def content_type(self) -> str:
+        """The MIME type of the data sent to the inference server."""
+
+    @property
+    @abc.abstractmethod
+    def accept(self) -> Tuple[str]:
+        """The content type(s) that are expected from the inference server."""
+
+
+class Predictor(PredictorBase):
     """Make prediction requests to an Amazon SageMaker endpoint."""
 
     def __init__(

src/sagemaker/serverless/__init__.py

@@ -0,0 +1,15 @@
+# Copyright 2017-2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""Classes for performing machine learning on serverless compute."""
+from sagemaker.serverless.model import LambdaModel  # noqa: F401
+from sagemaker.serverless.predictor import LambdaPredictor  # noqa: F401

src/sagemaker/serverless/model.py

@@ -0,0 +1,90 @@
+# Copyright 2017-2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""Models that can be deployed to serverless compute."""
+from __future__ import absolute_import
+
+import time
+from typing import Optional
+
+import boto3
+import botocore
+
+from sagemaker.model import ModelBase
+
+from .predictor import LambdaPredictor
+
+
+class LambdaModel(ModelBase):
+    """A model that can be deployed to Lambda."""
+
+    def __init__(
+        self, image_uri: str, role: str, client: Optional[botocore.client.BaseClient] = None
+    ) -> None:
+        """Initialize instance attributes.
+
+        Arguments:
+            image_uri: URI of a container image in the Amazon ECR registry. The image
+                should contain a handler that performs inference.
+            role: The Amazon Resource Name (ARN) of the IAM role that Lambda will assume
+                when it performs inference
+            client: The Lambda client used to interact with Lambda.
+        """
+        self._client = client or boto3.client("lambda")
+        self._image_uri = image_uri
+        self._role = role
+
+    def deploy(
+        self, function_name: str, timeout: int, memory_size: int, wait: bool = True
+    ) -> LambdaPredictor:
+        """Create a Lambda function using the image specified in the constructor.
+
+        Arguments:
+            function_name: The name of the function.
+            timeout: The number of seconds that the function can run for before being terminated.
+            memory_size: The amount of memory in MB that the function has access to.
+            wait: If true, wait until the deployment completes (default: True).
+
+        Returns:
+            A LambdaPredictor instance that performs inference using the specified image.
+        """
+        response = self._client.create_function(
+            FunctionName=function_name,
+            PackageType="Image",
+            Role=self._role,
+            Code={
+                "ImageUri": self._image_uri,
+            },
+            Timeout=timeout,
+            MemorySize=memory_size,
+        )
+
+        if not wait:
+            return LambdaPredictor(function_name, client=self._client)
+
+        # Poll function state.
+        polling_interval = 5
+        while response["State"] == "Pending":
+            time.sleep(polling_interval)
+            response = self._client.get_function_configuration(FunctionName=function_name)
+
+        if response["State"] != "Active":
+            raise RuntimeError("Failed to deploy model to Lambda: %s" % response["StateReason"])
+
+        return LambdaPredictor(function_name, client=self._client)
+
+    def delete_model(self) -> None:
+        """Destroy resources associated with this model.
+
+        This method does not delete the image specified in the constructor. As
+        a result, this method is a no-op.
+        """

src/sagemaker/serverless/predictor.py

@@ -0,0 +1,81 @@
+# Copyright 2017-2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""Predictors that are hosted on serverless compute."""
+from __future__ import absolute_import
+
+from typing import Optional, Tuple
+
+import boto3
+import botocore
+
+from sagemaker import deserializers, serializers
+from sagemaker.predictor import PredictorBase
+
+
+class LambdaPredictor(PredictorBase):
+    """A deployed model hosted on Lambda."""
+
+    def __init__(
+        self, function_name: str, client: Optional[botocore.client.BaseClient] = None
+    ) -> None:
+        """Initialize instance attributes.
+
+        Arguments:
+            function_name: The name of the function.
+            client: The Lambda client used to interact with Lambda.
+        """
+        self._client = client or boto3.client("lambda")
+        self._function_name = function_name
+        self._serializer = serializers.JSONSerializer()
+        self._deserializer = deserializers.JSONDeserializer()
+
+    def predict(self, data: dict) -> dict:
+        """Invoke the Lambda function specified in the constructor.
+
+        This function is synchronous. It will only return after the function
+        has produced a prediction.
+
+        Arguments:
+            data: The data sent to the Lambda function as input.
+
+        Returns:
+            The data returned by the Lambda function.
+        """
+        response = self._client.invoke(
+            FunctionName=self._function_name,
+            InvocationType="RequestResponse",
+            Payload=self._serializer.serialize(data),
+        )
+        return self._deserializer.deserialize(
+            response["Payload"],
+            response["ResponseMetadata"]["HTTPHeaders"]["content-type"],
+        )
+
+    def delete_endpoint(self) -> None:
+        """Destroy the Lambda function specified in the constructor."""
+        self._client.delete_function(FunctionName=self._function_name)
+
+    @property
+    def content_type(self) -> str:
+        """The MIME type of the data sent to the Lambda function."""
+        return self._serializer.CONTENT_TYPE
+
+    @property
+    def accept(self) -> Tuple[str]:
+        """The content type(s) that are expected from the Lambda function."""
+        return self._deserializer.ACCEPT
+
+    @property
+    def function_name(self) -> str:
+        """The name of the Lambda function this predictor invokes."""
+        return self._function_name

src/sagemaker/workflow/step_collections.py

@@ -136,10 +136,8 @@ def __init__(
         elif model is not None:
             if isinstance(model, PipelineModel):
                 self.model_list = model.models
-                self.container_def_list = model.pipeline_container_def(inference_instances[0])
             elif isinstance(model, Model):
                 self.model_list = [model]
-                self.container_def_list = [model.prepare_container_def(inference_instances[0])]
 
             for model_entity in self.model_list:
                 if estimator is not None:
@@ -154,10 +152,10 @@ def __init__(
                     source_dir = model_entity.source_dir
                     dependencies = model_entity.dependencies
                     kwargs = dict(**kwargs, output_kms_key=model_entity.model_kms_key)
-                    name = model_entity.name or model_entity._framework_name
+                    model_name = model_entity.name or model_entity._framework_name
 
                     repack_model_step = _RepackModelStep(
-                        name=f"{name}RepackModel",
+                        name=f"{model_name}RepackModel",
                         depends_on=depends_on,
                         sagemaker_session=sagemaker_session,
                         role=role,
@@ -171,10 +169,14 @@ def __init__(
                     model_entity.model_data = (
                         repack_model_step.properties.ModelArtifacts.S3ModelArtifacts
                     )
-
                     # remove kwargs consumed by model repacking step
                     kwargs.pop("output_kms_key", None)
 
+            if isinstance(model, PipelineModel):
+                self.container_def_list = model.pipeline_container_def(inference_instances[0])
+            elif isinstance(model, Model):
+                self.container_def_list = [model.prepare_container_def(inference_instances[0])]
+
         register_model_step = _RegisterModelStep(
             name=name,
             estimator=estimator,

tests/data/serverless/Dockerfile

@@ -0,0 +1,12 @@
+FROM public.ecr.aws/lambda/python:3.8
+
+COPY requirements.txt /tmp/requirements.txt
+RUN pip3 install -r /tmp/requirements.txt
+
+COPY model.py .
+RUN python model.py
+
+COPY app.py .
+COPY classes.txt .
+
+CMD ["app.handler"]

tests/data/serverless/README.md

@@ -0,0 +1,6 @@
+This folder contains the source code for the image used in the
+`sagemaker.serverless` tests.
+
+The image contains the code for a Lambda handler and all of its dependencies.
+The Lambda handler predicts the class of an image using a pre-trained ResNet-34
+model.

tests/data/serverless/app.py

@@ -0,0 +1,34 @@
+import urllib
+import json
+import os
+
+import torch
+from PIL import Image
+from torchvision import models
+from torchvision import transforms
+
+transform = transforms.Compose(
+    [
+        transforms.Resize(256),
+        transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
+    ]
+)
+
+with open("classes.txt") as file:
+    classes = [s.strip() for s in file.readlines()]
+
+
+def handler(event, context):
+    data = urllib.request.urlopen(event["url"])
+
+    image = Image.open(data)
+    image = transform(image)
+    image = image.unsqueeze(0)
+
+    model = torch.jit.load("./model.pt")
+    outputs = model(image)
+    target = outputs.argmax().item()
+
+    return {"class": classes[target]}