The azure.ai.ml.entities.SynapseSparkCompute class constructor in the Azure Machine Learning SDK v2 is used to define and attach an existing Azure Synapse Spark pool as a compute target. The constructor requires two primary arguments: name, which is a user-defined name to identify the compute target within the Azure Machine Learning workspace, and resourceid, which is the full Azure Resource Manager (ARM) ID of the specific Synapse Spark pool resource. This ARM ID uniquely identifies the pool within your Azure subscription.

A. The constructor requires the full resource ID, not a web URL and the pool name as separate arguments.

C. The name for the compute target is user-defined, not system-assigned, and a resource ID is needed, not a pool URL.

D. The Synapse workspace name and its URL are not the direct parameters required to attach a specific Spark pool compute target.

---

1. Microsoft. (2023). azure.ai.ml.entities.SynapseSparkCompute class. Azure Machine Learning SDK Documentation. Retrieved from https://learn.microsoft.com/en-us/python/api/azure-ai-ml/azure.ai.ml.entities.synapsesparkcompute?view=azure-python.

Reference Specifics: The "Constructor" section explicitly lists the parameters as name: str and resourceid: str.

2. Microsoft. (2023). Attach and use Synapse Spark pools in Azure Machine Learning (preview). Azure Machine Learning Documentation. Retrieved from https://learn.microsoft.com/en-us/azure/machine-learning/how-to-attach-synapse-spark-pool?view=azureml-api-2&tabs=sdk.

Reference Specifics: Under the "Attach a pool with the SDK" section

the example code demonstrates the usage: synapsecomputetarget = SynapseSparkCompute(name="synapse-spark"

resourceid="/subscriptions//...")

confirming the need for a user-defined name and the full resourceid.

When submitting a command job to an Azure Kubernetes Service (AKS) compute target using the Azure ML Python SDK v2, the instancetype parameter is used to specify the virtual machine (VM) size for the job. If the AKS cluster is configured with multiple node pools, each with a different VM size, specifying the instancetype directs the Azure ML scheduler to run the job on a node within the corresponding node pool. This allows for the selection of different compute node types for different workloads as required by the question.

A. compute: This parameter specifies the name of the compute target (e.g., the AKS cluster 'AKS1') where the job will run, not the specific type of node within that cluster.

B. environment: This parameter defines the software environment, including the Docker image and package dependencies for the job, not the underlying hardware infrastructure.

D. limits: This parameter is used to set resource constraints (like CPU or memory limits) for the container running the job, but it does not select the underlying VM node type.

1. Microsoft Official Documentation - azure.ai.ml.command function: The signature for the command function clearly lists instancetype as a parameter. The description states it is "The instance type of the compute target."

Source: Microsoft

Azure Machine Learning SDK documentation

"azure.ai.ml package - command function".

2. Microsoft Official Documentation - Train models with Azure Machine Learning CLI and Python SDK (v2): In the "Configure the job" section for a command job

the documentation shows an example of setting the instancetype within the job definition to specify the VM size for the run.

Source: Microsoft

Azure Machine Learning documentation

"Train models (create jobs)".

3. Microsoft Official Documentation - Attach and manage Kubernetes clusters: This document explains how instancetype is used with Kubernetes compute. It states

"When you submit a job to a Kubernetes compute target

you can use the instancetype parameter to specify the VM size for the job to run on. If the AKS cluster has a node pool with the specified VM size

the job will be scheduled to a node in that node pool."

Source: Microsoft

Azure Machine Learning documentation

"Configure and attach Kubernetes clusters in Azure Machine Learning"

Section: "Use instance type for job targeting".

The allowreuse parameter in an Azure Machine Learning pipeline step definition controls whether the step's output can be cached and reused. When allowreuse is set to True (which is the default behavior), the Azure Machine Learning service calculates a hash based on the step's inputs, parameters, and source code. For a PythonScriptStep, this hash includes the contents of the sourcedirectory. If a subsequent pipeline run is triggered with no changes to these components, the service finds a matching hash and reuses the output from the previous run instead of executing the step again. Therefore, including allowreuse=True ensures the desired caching behavior based on the sourcedirectory's content.

B. hashpaths: This parameter is used to specify additional files or directories to be included in the hash calculation. The sourcedirectory is already hashed by default, so this is not required for the described scenario.

C. dataasinput(name=): This is a method used to define data inputs for a step. While changes to inputs affect reuse, this option does not control the caching mechanism itself based on source code.

D. version: This parameter is typically used for manual versioning of components or environments. It does not enable automatic reuse based on unchanged source directory content.

1. Microsoft. (2023). azureml.pipeline.steps.PythonScriptStep class. Azure Machine Learning SDK for Python Documentation. Retrieved from https://learn.microsoft.com/en-us/python/api/azureml-pipeline-steps/azureml.pipeline.steps.pythonscriptstep?view=azure-ml-py.

Details: The documentation for the allowreuse parameter states

"Indicates whether the step should reuse previous results when re-run with the same settings. Reuse is enabled by default." The documentation for hashpaths clarifies that by default

"the contents of the sourcedirectory are hashed."

2. Microsoft. (2023). Troubleshoot machine learning pipelines - Reuse and triggers. Azure Machine Learning Documentation. Retrieved from https://learn.microsoft.com/en-us/azure/machine-learning/how-to-troubleshoot-pipelines?view=azureml-api-1#reuse-and-triggers.

Details: This document explains that the hash for determining reuse is based on several factors

including "The contents of the sourcedirectory" and explicitly states

"You can force a step to rerun by setting allowreuse=False in the step definition

" confirming that allowreuse is the primary control for this feature.

The scenario requires a compute target for a scheduled, batch inference pipeline. An Azure Machine Learning compute cluster (AmlCompute) is the ideal choice for this purpose. It is a managed compute resource specifically designed for running training or batch inference workloads. It can automatically scale up from zero nodes when a pipeline job is submitted and scale down when the job is complete, making it a cost-effective solution for recurring, scheduled tasks like a nightly inference run.

A. Azure Machine Learning compute instance: This is a single-node, managed workstation primarily intended for interactive development and authoring, not for running automated, production-level batch pipelines.

C. Azure Kubernetes Service (AKS)-based inference cluster: AKS is designed for deploying models as high-availability, scalable real-time web services, which handle immediate requests, not for executing scheduled batch processing pipelines.

D. Azure Container Instance (ACI) compute target: ACI is used for deploying models as real-time web services for low-scale or development/testing scenarios. It is not suitable for scalable, scheduled batch inference.

---

1. Microsoft Documentation | What are compute targets in Azure Machine Learning?

Section: Compute targets for training

Content: This document explicitly states

"The primary use case for a compute cluster is for training a machine learning model

but you can also use it for batch inference." It contrasts this with compute instances (for development) and inference clusters like AKS/ACI (for real-time deployment).

2. Microsoft Documentation | What are Azure Machine Learning pipelines?

Section: Which Azure pipeline should I use?

Content: This page clarifies that Azure Machine Learning pipelines are used for batch scoring scenarios. It further explains that pipeline steps run on compute targets

with compute clusters being the standard for scalable

automated jobs.

3. Microsoft Documentation | Create and run machine learning pipelines by using the Azure Machine Learning SDK

Section: Build the pipeline

Content: The official tutorials and examples for building pipelines consistently demonstrate setting a compute cluster as the compute target for the pipeline steps

reinforcing its role as the designated resource for such tasks. For example

job.compute = "cpu-cluster".

The solution is incorrect because it proposes using an Azure Kubernetes Service (AKS) cluster for a training job. The scenario explicitly states that the aks-cluster was created as an inference compute target. In Azure Machine Learning, compute targets are designated for specific purposes. AKS is the primary, recommended compute target for deploying models as real-time web services (inference), not for running training experiments. The appropriate compute target for training a deep neural network is an Azure Machine Learning compute cluster (AmlCompute), which is a managed service optimized for scalable training workloads. Using a compute target configured for inference to perform training is a fundamental misuse of the resource.

A. Yes: This is incorrect. The solution fails because it uses a compute resource (AKS for inference) for a task (training) for which it is not designed or intended within the Azure Machine Learning framework.

1. Microsoft Azure Documentation

"What are compute targets in Azure Machine Learning?": This official document explicitly categorizes compute targets. Azure Kubernetes Service is listed under "Compute targets for deployment" (inference)

while Azure Machine Learning compute clusters (AmlCompute) are listed under "Training compute targets". This distinction is fundamental to the service's design.

See section: "Training compute targets" and "Compute targets for deployment".

2. Microsoft Azure Documentation

"Train models with Azure Machine Learning": This guide details the process for training models. It consistently demonstrates the creation and use of an Azure Machine Learning compute cluster (AmlCompute) as the compute target for submitting a training job (ScriptRunConfig or command). It does not recommend or show AKS as a target for training runs.

See section: "Create a compute target".

3. Microsoft Azure Documentation

"Deploy a machine learning model to an Azure Kubernetes Service cluster": This document outlines the workflow for using AKS. The entire process described—creating an inference configuration and a deployment configuration—is centered on deploying a trained model for real-time scoring

confirming its role as an inference target.

See section: "Deploy to AKS".

The provided Python code correctly uses the Workspace.get() static method from the azureml.core library. This method is designed to retrieve an existing workspace object. The code supplies all the necessary parameters to uniquely identify the target workspace: the workspace name ('ml-project'), the subscriptionid ('sub1'), and the resourcegroup ('rg1'). These values precisely match the location of the 'ml-project' workspace as depicted in the diagram. Therefore, the code will successfully obtain a reference to the specified workspace.

B. No: This option is incorrect. The code provides the exact, correct parameters required by the Workspace.get() method to locate and connect to the specified workspace. There are no errors in the syntax or logic.

1. Microsoft Azure Machine Learning Documentation

azureml.core.workspace.Workspace class: "The get method retrieves a workspace object. It requires the workspace name

and either the subscription ID and resource group name

or a configuration file (config.json)." The provided solution uses the explicit parameter method.

Source: Microsoft

Azure Machine Learning SDK for Python documentation

Workspace Class

get method. Available at: https://learn.microsoft.com/en-us/python/api/azureml-core/azureml.core.workspace.workspace?view=azure-ml-py#azureml-core-workspace-workspace-get

2. Microsoft Learn

"Manage Azure Machine Learning workspaces": This tutorial demonstrates connecting to a workspace using the Python SDK. The example code provided is ws = Workspace.get(name="myworkspace"

subscriptionid=''

resourcegroup='myresourcegroup')

which is structurally identical to the solution in the question.

Source: Microsoft Learn

How to manage workspace

"Connect to a workspace" section. Available at: https://learn.microsoft.com/en-us/azure/machine-learning/how-to-manage-workspace?tabs=python#connect-to-a-workspace

The Import Data module in Azure Machine Learning Designer is specifically designed to load data from various external sources directly into a pipeline. It supports ingesting data from a "Web URL via HTTP," which perfectly matches the scenario of using a CSV file published on a website. This approach is the most direct and requires minimal administrative effort because it bypasses the need to first create and register a formal dataset asset in the Azure Machine Learning workspace.

A. Convert to CSV: This module is used for data output. It converts a dataset that is already inside the pipeline into CSV format, not for ingesting external data.

B. Enter Data Manually: This module is intended for creating very small datasets by typing or pasting data directly. It is not suitable for importing a file from a website.

D. Dataset: This refers to using a pre-registered dataset from your workspace's assets. The question explicitly states that a dataset has not been created for the file.

1. Microsoft Corporation. (2023). Import Data module. Microsoft Azure Documentation.

Reference: In the "How to configure Import Data" section

"Web URL via HTTP" is listed as a supported Data source. The document states

"Use the Import Data module to load data into a designer pipeline from existing cloud data services." This confirms its use for directly ingesting data from a URL.

2. Microsoft Corporation. (2023). Component and module reference. Microsoft Azure Documentation.

Reference: Under the "Data Input and Output" section

the description for Convert to CSV confirms it "Converts a dataset to a CSV format

" indicating it's an export/output function. The description for Enter Data Manually confirms it "Creates a small dataset by typing values." This documentation clearly distinguishes the purpose of these modules from data ingestion from a web source.

A Shared Access Signature (SAS) token is the most appropriate authentication method for this scenario. A SAS token provides secure, delegated access to resources in an Azure storage account. It allows for granular control by enabling you to specify permissions (such as read, write, list), a validity period (start and end time), and the specific resources (e.g., a specific container) that can be accessed. This directly addresses the requirement to "configure a privilege level" as the permissions are embedded within the token itself.

A. Account key: This provides full administrative access to the entire storage account and does not allow for configuring granular or restricted privilege levels.

C. Service principal: This is an identity. Its privileges are managed externally by assigning Azure RBAC roles (e.g., Storage Blob Data Reader), not configured within the credential itself.

D. Managed identity: Similar to a service principal, a managed identity is an identity whose access permissions are configured externally through Azure RBAC role assignments.

1. Microsoft Azure Documentation

"Create datastores": This document outlines the methods for creating datastores in Azure Machine Learning. It lists sastoken as a specific authentication parameter for registerazureblobcontainer

distinct from identity-based options. See the section "Authentication types".

2. Microsoft Azure Documentation

"Delegate access with a shared access signature": This source details how SAS tokens work. Under the section "How a SAS works

" it states

"With a SAS

you have granular control over how a client can access your data. For example: ... The permissions that are available to the client." This directly supports the concept of configuring a privilege level.

3. Microsoft Azure Documentation

"Connect to storage with identity-based data access": This document explains authentication using Azure Active Directory identities (service principals and managed identities). It clarifies that for these methods

you "grant your managed identity access to your storage account" by assigning an appropriate RBAC role

confirming that privilege configuration is external to the identity itself. See the section "Grant access to the storage service".

The most efficient way to grant an Azure Machine Learning workspace access to data in an existing Azure storage account, while minimizing data movement, is to register that storage account as a datastore. A datastore acts as a secure reference or pointer to the storage service, abstracting the connection details. This process does not copy or move the 100,000 image files; it simply establishes a connection, allowing Azure ML services to read the data directly from its original location during training. This approach directly satisfies the core requirement of minimizing data movement.

A. This option violates the constraint to minimize data movement, as it requires moving all photographs to a new storage service.

B. Azure Cosmos DB is a NoSQL database service, not the appropriate or efficient choice for storing and accessing large image files for deep learning model training.

C. A TabularDataset is used for structured, tabular data (like from a CSV or Parquet file). For image files, a FileDataset should be used, making this option technically incorrect.

E. This option explicitly involves copying the data to the workspace's default storage, which directly contradicts the requirement to minimize data movement.

1. Microsoft Azure Documentation

"Create and manage datastores": This document states

"An Azure Machine Learning datastore is a reference to an existing storage account on Azure. A datastore abstracts the connection details... This allows you to securely access your storage without having to hardcode connection information in your scripts." This directly supports registering the existing storage as a datastore.

Source: Microsoft Corporation. (2023). Create and manage datastores in Azure Machine Learning. Microsoft Docs. Retrieved from https://docs.microsoft.com/en-us/azure/machine-learning/how-to-datastore

2. Microsoft Azure Documentation

"Create Azure Machine Learning datasets": This document clarifies the difference between dataset types. It specifies

"A FileDataset references single or multiple files in your datastores... The files can be of any format

which makes FileDataset a good choice for deep learning scenarios." This confirms that TabularDataset (Option C) is incorrect for image files.

Source: Microsoft Corporation. (2023). Create Azure Machine Learning datasets. Microsoft Docs. Retrieved from https://docs.microsoft.com/en-us/azure/machine-learning/how-to-create-register-datasets

3. Microsoft Azure Documentation

"Access data in Azure storage services": This guide outlines the recommended pattern for accessing data: "We recommend that you create an Azure Machine Learning datastore to store connection information to your Azure storage service. To create a datastore

you'll need... access to the supported storage type." This reinforces that creating a datastore is the standard and correct first step.

Source: Microsoft Corporation. (2023). Access data in Azure storage services. Microsoft Docs. Retrieved from https://docs.microsoft.com/en-us/azure/machine-learning/how-to-access-data

Azure Application Insights is the designated Azure service for monitoring live applications, including real-time inference services deployed via Azure Machine Learning. By enabling Application Insights for the service endpoint, you can automatically collect telemetry such as request rates, response times, and failure rates. You can also add custom logging to the scoring script to capture the specific request data and the model's predictions. This provides a comprehensive monitoring solution viewable in the Azure portal with minimal setup, as it's a direct integration that can be enabled with a single parameter during deployment.

A. View the explanations for the registered model in Azure ML studio.

Model explanations are for interpretability (understanding why a model makes a prediction), not for continuous operational monitoring of data traffic to a live endpoint.

C. Create an ML Flow tracking URI that references the endpoint, and view the data logged by ML Flow.

MLflow Tracking is primarily used for logging metrics and artifacts during the model training and experimentation phase, not for monitoring a deployed production endpoint.

D. View the log files generated by the experiment used to train the model.

Experiment logs contain information about the model training run. They are irrelevant to the live data being sent to the deployed inference service post-training.

---

1. Microsoft Documentation - Monitor Azure Machine Learning: "For data collected from a real-time endpoint

enable Azure Application Insights. This service collects application telemetry

such as request rates

response times

and failure rates. Application Insights can also be used to track custom events and data."

Source: Microsoft

"Monitor Azure Machine Learning

" Azure Machine Learning Documentation. (Found under the section "Monitor deployments").

2. Microsoft Documentation - How to monitor online endpoints: "Azure Machine Learning provides integration with Azure Application Insights to view metrics and create alerts for your online endpoints... You can enable or disable Application Insights when you create an online endpoint in Azure Machine Learning studio or by using the CLI and SDKs."

Source: Microsoft

"How to monitor online endpoints

" Azure Machine Learning Documentation. (Found in the "Monitor metrics with Application Insights" section).

3. Microsoft Documentation - Enable Application Insights for online endpoints: This document shows that enabling Application Insights can be done with a simple boolean flag (appinsightsenabled=True) in the deployment configuration

confirming it requires minimal administrative effort.

Source: Microsoft

"Enable Application Insights for online endpoints

" Azure Machine Learning Documentation. (Found in the "Enable with Python SDK" and "Enable with Azure CLI" sections).