AI Bot
AI-103: Azure Machine Learning Workspace & Pipeline Operations Flashcards
Microsoft Azure AI App and Agent Developer Associate AI-103 Flashcards
Study our AI-103: Azure Machine Learning Workspace & Pipeline Operations flashcards for the Microsoft Azure AI App and Agent Developer Associate AI-103 exam with 45+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| Best practices for reproducible runs | Pin environment versions register datasets and environments use pipeline step inputs outputs and log dependencies |
| CLI command to create a workspace | az ml workspace create followed by --name --resource-group and --location |
| Difference between Azure ML compute cluster and compute instance | Compute cluster is for scalable multi node training Compute instance is a development VM for interactive work |
| Difference between TabularDataset and FileDataset | TabularDataset is for structured row column data FileDataset is for files and folders |
| Differences between deployment targets ACI and AKS | ACI is for single instance testing and development AKS is for production scale low latency inference |
| How do you create a workspace using the Azure portal | Use the Azure portal new resource create Machine Learning workspace provide subscription resource group name and region then create |
| How do you stop or delete a compute instance | Stop or delete the compute resource from Azure portal or use SDK ComputeTarget.stop and ComputeTarget.delete |
| How does model versioning work | Each registered model name can have multiple versions each incremented automatically with registration |
| How to add tags and properties to runs or models | Use run.tag run.properties or Model.register tags and properties parameters to store metadata |
| How to attach an Azure Blob storage as a datastore | Use Datastore.register_azure_blob_container or via Azure portal provide container name account name and credentials |
| How to clean up workspace resources to avoid costs | Delete compute clusters instances endpoints published pipelines and optionally delete workspace from Azure portal |
| How to create a reusable environment | Create Environment object specify conda or pip dependencies register the environment in workspace |
| How to create an Azure ML compute cluster using SDK | Use AmlCompute.provisioning_configuration and ComputeTarget.create with min_nodes and max_nodes |
| How to debug failed pipeline steps | Examine step logs in run history download docker logs check stdoutstderr and inspect environment and dependencies |
| How to deploy a model to AKS | Create inference configuration deployment config then call Model.deploy targeting an existing or provisioned AKS cluster |
| How to log metrics from a training script | Use run.log or run.log_list from the azureml.core Run object inside the script |
| How to pass data between pipeline steps | Use PipelineData or OutputFileDatasetConfig shared artifacts stored in datastore accessible by subsequent steps |
| How to publish a pipeline | Call pipeline.publish to create a REST endpoint for triggering pipeline runs from external systems |
| How to register a dataset using SDK | Use Dataset.register_pandas_dataframe or TabularDatasetFactory.register or FileDatasetFactory.register with workspace and name |
| How to register a model from a run | Call Model.register with workspace model_path and name optionally providing description tags and run_id |
| How to restrict workspace access | Assign Azure RBAC roles and use resource policies private link and network isolation features |
| How to retrieve a model for deployment | Use Model(workspace name version) or Model.get_model_path to download model files in a run or script |
| How to start an AutoML run using SDK | Define AutoMLConfig with task settings featurization and compute then call AutoMLRun via Experiment.submit |
| How to submit a run in Azure ML | Define ScriptRunConfig or RunConfiguration then call Experiment.submit with the config and environment |
| How to track and compare model metrics across runs | Use Run history metrics and the RunDetails widget or use the registry tags and model properties for filtering |
| How to trigger a published pipeline on a schedule | Use Azure ML pipeline REST endpoint with Azure Logic Apps or Azure Data Factory or use schedule in ML studio |
| How to upload artifacts to a run | Use run.upload_file or run.upload_folder or log_model utilities within the run context |
| How to use managed identities with Azure ML | Assign workspace or compute managed identity grant it storage or keyvault roles for secure credentialless access |
| Name three core workspace artifacts | Datasets compute targets and models |
| What are common Pipeline step types | PythonScriptStep DataTransferStep ParallelRunStep and AutoMLStep |
| What is a datastore in Azure ML | A credentialed storage abstraction that points to Azure Blob ADLS or other storage and is used by datasets and compute |
| What is a Pipeline in Azure ML | Composable workflow of steps that orchestrates data prep training evaluation and registration with reproducible runs |
| What is a Run in Azure ML | An execution record of a submitted job that captures metrics logs outputs artifacts and system info |
| What is an AutoML experiment | An AutoML run that explores model pipelines with validation and returns best model candidates and metrics |
| What is an Azure ML compute target | An environment where training or inference jobs run such as compute clusters attached VMs or attached Kubernetes |
| What is an Azure ML dataset | A managed reference to data that supports versioning access controls and convenience functions for training |
| What is an Azure ML workspace | Central Azure resource that organizes experiments datasets compute models deployments and access control |
| What is an Environment in Azure ML | Defines software dependencies runtime and Docker image used to run training and inference |
| What is an Experiment in Azure ML | Logical container for runs that groups related training or evaluation jobs for tracking and comparison |
| What is an inference config | Specifies entry script environment and target framework used during model deployment for scoring |
| What is an OutputFileDatasetConfig | Declarative output dataset object that registers step outputs as a dataset and materializes to datastore |
| What is AutoML in Azure ML | Automated machine learning service that selects algorithms hyperparameters and preprocessing to produce candidate models |
| What is dataset consumption mode mount versus download | Mount streams files from datastore at runtime download copies files into compute local storage choose based on performance and isolation |
| What is node autoscale setting for AmlCompute | Enables automatic scaling of target node count between min_nodes and max_nodes based on job demand |
| What is the Model Registry in Azure ML | A workspace scoped registry of registered models with versions metadata and associated artifacts |
About the Flashcards
Flashcards for the Microsoft Azure AI App and Agent Developer Associate exam help students review core Azure Machine Learning terminology, concepts, and common workflows. Cards explain workspace roles and core artifacts (datasets, compute targets, models, deployments, and access control), dataset types and datastores, and resource creation via portal, CLI, and SDK.
Focused cards cover experiments and runs (submission, logging, artifacts), model registry and versioning, environments and inference configs, deployment targets and procedures (ACI vs AKS), pipelines and AutoML workflows, plus operational topics like managed identities, debugging failed steps, reproducible runs, access restrictions, and resource cleanup to control costs.
Topics covered in this flashcard deck:
- Azure ML workspace
- Datasets and datastores
- Compute targets and autoscale
- Experiments, runs, logging
- Model registry and deployment
- Pipelines and AutoML