Explore Data and Run Experiments:
Sample Questions and Answers¶

Note

The questions and answers provided in this study guide are for practice purposes only and are not official practice questions. They are intended to help you prepare for the DP-100 Microsoft certification exam. For additional preparation materials and the most up-to-date information, please refer to the official Microsoft documentation.

Topics Covered¶

Explore and prepare data
Create and run experiments
Track experiments with MLflow
Use Automated Machine Learning (AutoML)

List of References (Click to expand)

List of questions/answers (Click to expand)

Q1: Data Asset Types
Q2: MLflow Logging Functions
Q3: AutoML Configuration
Q4: Command Job Parameters
Q5: Data Asset Paths
Q6: AutoML Data Guardrails
Q7: Experiment Tracking
Q8: Data Access Methods
Q9: Environment Management
Q10: AutoML Limits

Tip

Data Asset Types:

Asset Type	Use Case	Best For
FileDataset	Raw files, images, documents	Unstructured data, computer vision tasks
TabularDataset	Structured row-column data	Traditional ML, CSV/Parquet files
mltable	Standardized dataset definition	v2 SDK, AutoML, reproducible pipelines
uri_folder	Reference to storage folder	Batch processing, file collections

Tip

Supported Data Paths:

Path Type	Format	Example
Local	`./<path>`	`./data/train.csv`
Azure Blob	`wasbs://`	`wasbs://account.blob.core.windows.net/container/file`
Azure Data Lake Gen2	`abfss://`	`abfss://filesystem@account.dfs.core.windows.net/folder/file`
Datastore	`azureml://`	`azureml://datastores/datastore_name/paths/folder/file`

Tip

MLflow Logging Functions:

Function	Purpose	Example Usage
mlflow.log_param()	Log single parameter	`mlflow.log_param("learning_rate", 0.01)`
mlflow.log_metric()	Log single metric	`mlflow.log_metric("accuracy", 0.95)`
mlflow.log_artifact()	Log file (plot, model)	`mlflow.log_artifact("confusion_matrix.png")`
mlflow.log_model()	Log ML model	`mlflow.sklearn.log_model(model, "model")`

Tip

AutoML Experiment Limits:

Parameter	Description	Impact
timeout_minutes	Total experiment duration	Prevents runaway experiments
trial_timeout_minutes	Maximum time per model trial	Controls individual model training time
max_trials	Maximum number of models to train	Balances exploration vs time
enable_early_termination	Stop if score isn't improving	Saves compute resources

Tip

AutoML Data Guardrails:

Guardrail	Purpose	States
Class Balancing Detection	Identifies imbalanced datasets	Passed / Done / Alerted
Missing Feature Values Imputation	Handles missing data	Passed / Done / Alerted
High Cardinality Feature Detection	Identifies features with many unique values	Passed / Done / Alerted

Tip

Command Job Configuration:

Parameter	Description	Required
code	Folder containing the script	✅ Yes
command	Script to execute	✅ Yes
environment	Runtime environment	✅ Yes
compute	Compute target	✅ Yes
display_name	Job display name	❌ Optional
experiment_name	Experiment grouping	❌ Optional

Q1: Data Asset Types¶

You need to work with a collection of image files for a computer vision project. Which data asset type should you use?

TabularDataset ❌: Incorrect. TabularDataset is for structured, row-column data.
mltable ❌: Incorrect. mltable is primarily for structured data with schema.
FileDataset ✅: Correct. FileDataset is designed for working with raw files like images.
uri_folder ❌: Incorrect. While possible, FileDataset provides better functionality for file collections.

Q2: MLflow Logging Functions¶

You want to log the learning rate hyperparameter and accuracy metric for your experiment. Which MLflow functions should you use?

mlflow.log_artifact() for both ❌: Incorrect. log_artifact() is for files, not parameters or metrics.
mlflow.log_param() for learning rate, mlflow.log_metric() for accuracy ✅: Correct. log_param() for hyperparameters, log_metric() for performance metrics.
mlflow.log_metric() for both ❌: Incorrect. Parameters should use log_param().
mlflow.log_model() for both ❌: Incorrect. log_model() is specifically for ML models.

Q3: AutoML Configuration¶

You want to prevent AutoML experiments from running indefinitely. Which parameter should you configure?

max_trials ❌: Incorrect. This limits the number of models, not time.
trial_timeout_minutes ❌: Incorrect. This limits individual trial time, not total experiment time.
timeout_minutes ✅: Correct. This parameter sets the maximum duration for the entire AutoML experiment.
enable_early_termination ❌: Incorrect. This helps stop underperforming trials, but doesn't set a time limit.

Q4: Command Job Parameters¶

Which parameters are required when configuring a command job in Azure ML? (Select all that apply)

code ✅: Correct. The folder containing the script is required.
command ✅: Correct. The command to execute is required.
environment ✅: Correct. The runtime environment must be specified.
compute ✅: Correct. The compute target must be specified.
display_name ❌: Incorrect. Display name is optional.

Q5: Data Asset Paths¶

Which path format should you use to reference a file in Azure Data Lake Storage Gen2?

wasbs://account.blob.core.windows.net/container/file ❌: Incorrect. This format is for Azure Blob Storage.
abfss://filesystem@account.dfs.core.windows.net/folder/file ✅: Correct. This is the correct format for Azure Data Lake Storage Gen2.
azureml://datastores/name/paths/file ❌: Incorrect. This format is for registered datastores.
./data/file.csv ❌: Incorrect. This format is for local files.

Q6: AutoML Data Guardrails¶

What are the three possible states for AutoML data guardrails?

Passed, Failed, Warning ❌: Incorrect. These are not the correct state names.
Passed, Done, Alerted ✅: Correct. These are the three states for data guardrails.
Success, Error, Pending ❌: Incorrect. These are not the data guardrail states.
Complete, Incomplete, Skipped ❌: Incorrect. These are not the correct states.

Q7: Experiment Tracking¶

You want to track multiple related machine learning experiments. What should you do?

Create a separate workspace for each experiment ❌: Incorrect. Workspaces are for broader organizational boundaries.
Use the same experiment name for all related runs ✅: Correct. Experiment names group related runs together for comparison.
Create a new MLClient for each experiment ❌: Incorrect. One MLClient can handle multiple experiments.
Use different compute targets for each experiment ❌: Incorrect. Compute targets don't affect experiment organization.

Q8: Data Access Methods¶

You have a large TabularDataset and want to process it with Spark. Which method should you use?

data.to_pandas_dataframe() ❌: Incorrect. Pandas is for in-memory processing, not suitable for large datasets.
data.to_spark_dataframe() ✅: Correct. This converts the dataset to Spark DataFrame for distributed processing.
data.get() ❌: Incorrect. This downloads files locally, not suitable for Spark processing.
data.sample() ❌: Incorrect. This only returns a sample, not the full dataset.

Q9: Environment Management¶

What is the best practice for managing environments across multiple experiments?

Create a new environment for each experiment ❌: Incorrect. This creates unnecessary duplication.
Create and register an environment, then reuse it ✅: Correct. Registered environments ensure reproducibility and reduce maintenance.
Always use curated environments ❌: Incorrect. Curated environments may not have all required packages.
Share Dockerfile manually between team members ❌: Incorrect. This doesn't leverage Azure ML's environment management.

Q10: AutoML Limits¶

You want to train a maximum of 50 models in your AutoML experiment, with each model taking no more than 15 minutes. How should you configure the limits?

max_trials=50, timeout_minutes=15 ❌: Incorrect. timeout_minutes is for the entire experiment, not individual trials.
max_trials=50, trial_timeout_minutes=15 ✅: Correct. max_trials limits the number of models, trial_timeout_minutes limits each individual trial.
timeout_minutes=50, trial_timeout_minutes=15 ❌: Incorrect. timeout_minutes=50 would limit the entire experiment to 50 minutes.
max_trials=15, timeout_minutes=50 ❌: Incorrect. This reverses the intended limits.

Code Examples¶

Creating and Logging to an Experiment¶

import mlflow
import mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score

# Set experiment
mlflow.set_experiment("customer-churn-prediction")

with mlflow.start_run():
    # Log parameters
    mlflow.log_param("n_estimators", 100)
    mlflow.log_param("max_depth", 10)

    # Train model
    model = RandomForestClassifier(n_estimators=100, max_depth=10)
    model.fit(X_train, y_train)

    # Make predictions and log metrics
    predictions = model.predict(X_test)
    accuracy = accuracy_score(y_test, predictions)
    mlflow.log_metric("accuracy", accuracy)

    # Log model
    mlflow.sklearn.log_model(model, "model")

Configuring a Command Job¶

from azure.ai.ml import command
from azure.ai.ml.entities import Environment

# Define environment
env = Environment(
    name="sklearn-env",
    conda_file="environment.yml",
    image="mcr.microsoft.com/azureml/minimal-ubuntu20.04-py38-cpu-inference:latest"
)

# Configure command job
job = command(
    code="./src",
    command="python train.py --data-path ${{inputs.training_data}} --model-name ${{inputs.model_name}}",
    inputs={
        "training_data": Input(type="uri_folder", path="azureml://datastores/workspaceblobstore/paths/data/"),
        "model_name": "random-forest"
    },
    environment=env,
    compute="cpu-cluster",
    experiment_name="model-training"
)

# Submit job
ml_client.jobs.create_or_update(job)

AutoML Configuration¶

from azure.ai.ml import automl

# Configure AutoML classification job
classification_job = automl.classification(
    compute="cpu-cluster",
    experiment_name="automl-classification",
    training_data=training_data,
    target_column_name="target",
    primary_metric="accuracy",
    n_cross_validations=5,
    enable_model_explainability=True,
    timeout_minutes=60,
    trial_timeout_minutes=20,
    max_trials=10,
    enable_early_termination=True
)

# Submit AutoML job
ml_client.jobs.create_or_update(classification_job)

Best Practices¶

Data Management:
Use versioned datasets for reproducibility
Validate data quality before training
Monitor for data drift in production
Experiment Tracking:
Use descriptive experiment and run names
Log hyperparameters, metrics, and artifacts consistently
Tag runs with relevant metadata
AutoML Optimization:
Set appropriate timeout limits to control costs
Enable early termination for efficiency
Review data guardrails and address issues
Environment Management:
Register and version environments
Use minimal base images for faster startup
Test environments before production use

Explore Data and Run Experiments: Sample Questions and Answers¶

Topics Covered¶

Q1: Data Asset Types¶

Q2: MLflow Logging Functions¶

Q3: AutoML Configuration¶

Q4: Command Job Parameters¶

Q5: Data Asset Paths¶

Q6: AutoML Data Guardrails¶

Q7: Experiment Tracking¶

Q8: Data Access Methods¶

Q9: Environment Management¶

Q10: AutoML Limits¶

Code Examples¶

Creating and Logging to an Experiment¶

Configuring a Command Job¶

AutoML Configuration¶

Best Practices¶

Explore Data and Run Experiments:
Sample Questions and Answers¶