Microsoft Azure Data Scientist Associate (DP-100): Essential Concepts

The Azure Data Scientist Associate certification (DP-100) validates applying data science and ML techniques with Azure Machine Learning. This guide covers the concepts that show up most often in scenario questions.

1. Azure Machine Learning Workspace#

The Azure ML workspace is your central hub for all machine learning activities. Understanding its components is fundamental to the exam.

Core Workspace Components#

Compute Options Compared#

Workspace Security#

• Managed Identity — Azure ML can use managed identities to access resources without storing credentials
• Private Link — Keep all traffic within your virtual network
• Role-Based Access Control — Fine-grained permissions for workspace resources
• Customer-Managed Keys — Encrypt workspace data with your own keys

Test Your Understanding#

2. Data Management & Feature Engineering#

Effective data management is critical for reproducible ML workflows. Azure ML provides robust tools for versioning and tracking data.

Data Assets#

Data Assets in Azure ML provide:

• Versioning — Track changes to your datasets over time
• Lineage — Know exactly which data version trained each model
• Reproducibility — Recreate any experiment with the exact same data
• Sharing — Collaborate across teams with consistent data access

from azure.ai.ml.entities import Data
from azure.ai.ml.constants import AssetTypes

# Create a versioned data asset
my_data = Data(
    name="customer-churn-data",
    version="1.0",
    path="azureml://datastores/blob_store/paths/data/churn.csv",
    type=AssetTypes.URI_FILE,
    description="Customer churn dataset for Q4 2025"
)
ml_client.data.create_or_update(my_data)

Feature Engineering Best Practices#

Handling Class Imbalance#

Class imbalance is a common exam topic. Key techniques:

• SMOTE — Synthetic Minority Over-sampling Technique
• Class Weights — Adjust algorithm to penalize minority class errors more
• Stratified Sampling — Maintain class proportions in train/test splits
• Undersampling — Reduce majority class (use with caution)

3. Model Training & Hyperparameter Tuning#

Azure ML provides powerful tools for training models at scale and optimizing hyperparameters.

Training with Jobs#

Azure ML jobs encapsulate your training logic and can run on any compute target:

from azure.ai.ml import command

# Define a training job
training_job = command(
    code="./src",
    command="python train.py --learning-rate ${{inputs.lr}} --epochs ${{inputs.epochs}}",
    inputs={
        "lr": 0.01,
        "epochs": 100
    },
    environment="AzureML-sklearn-1.0-ubuntu20.04-py38-cpu@latest",
    compute="gpu-cluster",
    experiment_name="churn-classification"
)

# Submit the job
ml_client.jobs.create_or_update(training_job)

Hyperparameter Tuning (Sweep Jobs)#

Early Termination Policies#

AutoML#

Azure AutoML automates model selection and hyperparameter tuning:

• Classification — Logistic regression, tree-based models, neural networks
• Regression — Linear regression, gradient boosting, neural networks
• Time Series Forecasting — ARIMA, Prophet, deep learning
• Computer Vision — Image classification, object detection
• NLP — Text classification, named entity recognition

4. Model Deployment & Endpoints#

Deploying models effectively is critical for delivering ML value to production applications.

Endpoint Types#

Managed Online Endpoints#

Online endpoints provide:

• Autoscaling — Scale based on CPU, memory, or request metrics
• Blue-green deployment — Traffic splitting between model versions
• Health monitoring — Built-in liveness and readiness probes
• Managed infrastructure — No cluster management required

from azure.ai.ml.entities import ManagedOnlineEndpoint, ManagedOnlineDeployment

# Create endpoint
endpoint = ManagedOnlineEndpoint(
    name="churn-prediction-endpoint",
    auth_mode="key"
)
ml_client.online_endpoints.begin_create_or_update(endpoint).result()

# Create deployment
deployment = ManagedOnlineDeployment(
    name="blue",
    endpoint_name="churn-prediction-endpoint",
    model=model,
    environment=env,
    code_configuration=CodeConfiguration(
        code="./src", scoring_script="score.py"
    ),
    instance_type="Standard_DS3_v2",
    instance_count=2
)
ml_client.online_deployments.begin_create_or_update(deployment).result()

Custom Environments#

For models with specific dependencies:

# conda.yaml
name: pytorch-inference
channels:
  - pytorch
  - conda-forge
dependencies:
  - python=3.9
  - pytorch=2.0
  - torchvision
  - pip:
    - azure-ai-ml
    - inference-schema

5. MLOps & Responsible AI#

Modern ML requires robust operational practices and ethical considerations.

MLflow Integration#

Azure ML integrates natively with MLflow for experiment tracking:

import mlflow

# Enable Azure ML as the MLflow backend
mlflow.set_tracking_uri(ml_client.tracking_uri)

# Log experiments
with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_metric("accuracy", 0.95)
    mlflow.sklearn.log_model(model, "model")

ML Pipelines#

Pipelines enable reproducible, automated workflows:

Model Monitoring & Drift Detection#

Production models degrade over time due to:

• Data Drift — Input data distribution changes
• Concept Drift — Relationship between features and target changes
• Model Degradation — Accuracy decreases over time

Azure ML provides:

• Data drift monitoring — Detect when production data differs from training data
• Model performance monitoring — Track prediction accuracy over time
• Automated alerts — Trigger retraining when thresholds are breached

Responsible AI Dashboard#

The Responsible AI dashboard provides:

---

Know your range before test day#

DP-100 rewards hands-on familiarity with Azure ML. Sit a short diagnostic, then practice free on the gaps before a timed module.

• Timed diagnostic with domain-level breakdown
• Clear signal on what to drill next
• Practice free; timed sittings and readiness range on Pro

Take the free DP-100 assessment →

---

Next step: Start practicing free on the DP-100 hub, or sit a short diagnostic first.