Topics Cover in Azure DS Cert.
( 4 )
Deploy & Operationalize ML Solutions (35 - 40%)
- Model deployment
Run Experiments & Train Models (20 - 25%)
- Build & Train Models
Manage Azure Resources for ML (25 - 30%)
- Set up & Manage an Azure Workspace
Implement Responsible ML (5 - 10%)
- Privacy, Feature Imp., Fairness
Subtopics: Deploy & Operationalize ML
(35 - 40%)
( 7 )
Understanding the genreal steps in deployment:
- select the compute, register the model, deploy the models, schedule runs/jobs, trigger a pipeline run
- perform these functions in SDK
- publish these as web services
Select Compute
- security for deployment
- evaluate compute options
Manage Models in Azure ML
- register & monitor models
- monitor data drift
Deploy Model as a Service
- configure deployment settings
- deploy models (Databricks built & other models)
- troubleshoot container issues
- consume deployed service
Create a Pipeline for Batch Inferencing (schedule runs)
- configure parallelRunStep, obtain outputs
- configure batch compute
- publish batch
- run batch pipeline
Apply ML Ops Practices
- trigger a pipeline
- automate model retraining with new data
- refactor notebooks into scripts & implement source controls
Publish Designer Pipeline as a Web Service
Implment Pipelines in SDK
Subtopics: Run & Train Models
(20 - 25 %)
( 5 )
Gerenal Overview:
- create, run, generate metrics, optimize, hyperparameter tuning
Create Models Using Designer
- create pipeline
- ingest data
- define data flow
- use custom code modules in designer
Run Model Training Scripts
- consume data & run with SDK
- configure run setting
- run script & train model in Databricks
Generate Metrics for Experiment Run
- log experiment metrics
- retrieve and view outputs
- troubleshoot run errors with logs
- use MLflow to track experiments
- track experiments in Databricks
Optimze Models w/ Automated ML
- use Automated ML interface in Azure Learning Studio
- use Automated ML in SDK
- pre-processing & algorithmn options
- primary metrics
- retrieve best model and get data from automated run
Hyperparameter Tuning
- select sampling method
- define search space & primary metric
- define early termination options
- define model with optimal hyperpatameter values
Subtopics: Manage Azure Resources for ML
(workspace stuff)
(25 - 30%)
( 6 )
Create Workspace
- creare, configure, & manage workspace
Manage Data in Workspace
- seelect storage resources and register datastores
- create & manage datasets
Manage Compute for Experiments
- determine appropriate compute ancreate compute targets
- configure & monitor compute
Implement Security & Access Controls
- manage & create roles
- manage credentials
- determine access requirements and map to user roles
Set Up Development Environment
- acess Azure ML from other development environments
- create & share compute instances (more other environ.)
Set Up Databricks Workspace
- create a DB workspace & cluster
- link DB workspace to Azure ML workspace
- create & run notesbooks
Subtopics: ML Ops
(5 - 10%)
( 3 )
Use Experiments to Interpret Models
- use model explainer
- generate feature importance data
Describe Fariness Cnsiderations for Models
- evaluate model fairness
- mitigate model unfairness
Describe Privacy Considerations for Data
- principals of differential privacy
- specify acceptable levels of noise in the data and ites effects on privacy
