DP-100 Azure Data Scientist Associate Complete Exam Guide

Create an Azure account, try Azure for free with $200 credit for 30 days, or choose pay-as-you-go, then proceed to create an Azure machine learning workspace.

Learn how an Azure machine learning workspace automatically provisions storage, application insights, key vault, and container registry, alongside compute clusters, notebooks, data registrations and assets, experiments, pipelines, and models.

Access Azure Machine Learning tools through four methods: Studio, the Python SDK, the CLI extension, and Visual Studio Code extensions, enabling visual, code-driven, and command-line workflows.

Create a compute instance in the workspace to run Python code and access Python SDK, using DS11 v2 with 2 cores, 14 GB RAM, 28 GB disk, at $0.18 hourly.

Stop compute instances when finished to avoid charges. Delete them afterward to keep resources clear and costs low.

Register and consume data assets using the sdk and notebooks, preview flight delay data in designer, convert datasets to pandas data frames, and perform basic transformations for analysis.

Load and transform flight delay data with data assets and data stores in Azure Machine Learning Designer, using prebuilt components, analyze data, apply differential privacy for model development.

Explore data using the data asset explorer and Azure Data Explorer, profiling top 10,000 rows and then ingesting a blob-stored parquet dataset, generating SAS URLs, and running queries.

Explore Azure Data Explorer queries with pipes to filter and project year, month, carrier, and departure delay; sort, summarize, and render a column chart for quick insights.

Profile the full flight delays dataset to explore data with histograms and box plots, identify outliers such as a 1863-minute department delay, and guide data cleaning decisions.

Ingest flight delay data into Azure Data Explorer, profile and explore arrival and departure delays by carrier and day of week, and build a model to predict arrival delay.

Prepare data for model training by clipping outliers with the 90th to 98th percentile and cleaning missing values, then select relevant flight delay features to build a ready training pipeline.

Prepare flight delay data, select features, add arrive delay as the target, split into 70/30, train a linear regression model in designer, and score predictions.

Create an automated ML regression model using a tabular data asset, predicting arrival delay from flight delay data. Configure compute options, run multiple trials, and identify the best model.

Practice automated ML regression on tabular data by adjusting timeouts to at least an hour to allow Azure to find the optimal model for flight delay data.

Learn automated ml regression workflow with data guardrails, feature imputation, and a stack ensemble model, exploring preprocessing, model metrics, and runnable notebook code.

Explore automated machine learning for regression and classification, learn how preprocessing, feature engineering, and model selection using the primary metric optimize models and validate with train/validation splits and test data.

Set up and run an experiment by importing experiments, creating an experiment, logging metrics (row count, carriers), converting data to a pandas dataframe, and comparing results across runs.

Prepare data in pandas, select day of month, day of week, and month to predict arrival delay, split into train and test sets with train_test_split (30% test, random state 42).

Train a linear regression model using the python sdk, fit it on x_train and y_train, evaluate with predictions on x_test, and log r-squared and mean squared error.

Run and compare experiments in Azure Machine Learning Studio, train and score models, log metrics like mean squared error and R squared, and exclude outliers to improve flight delay predictions.

Learn to create parameterized model training scripts that orchestrate data import, data preparation, and training, log metrics, while spinning up compute clusters and environments and using MLflow for tracking.

Configure an Azure ML environment with conda dependencies, including scikit-learn, pandas, and numpy, then set up a script run config and run the training script end-to-end.

Demonstrates running a model training script end-to-end by provisioning a new Azure ML environment, installing dependencies, executing the script, and saving the trained model to the uploads folder.

Launch end-to-end training by creating the script config and running in a fresh environment, using run details library to monitor errors and metrics (mean squared error, R squared).

Configure the training script parameters to run multiple experiments and log outputs for comparison. Tune ridge regression alpha for L2 regularization.

Learn to pass arguments to a training script using the ARG Pass module, define a float l2 rate parameter with a default of 0.0, and log results across multiple runs.

Cycle through script parameters by iterating different l2 rates, initialize script configs, and run experiments to compare outputs, noting how r squared and mean squared error respond.

Explore how changing the test size by adjusting the L2 rate affects model evaluation, comparing mean squared error and R squared across configurations with charts.

Attach a compute target to your training script in the run config, choosing either a compute target object, a target name, or local, and verify the short instance is used.

Spin up a compute cluster and attach it to a script run config to run training with the appropriate environment. Stop or delete the cluster afterward to avoid compute fees.

Learn how to build pipelines in Azure Machine Learning Studio using notebooks, orchestrating connected steps into cohesive machine learning workflows.

Create a pipeline run script to orchestrate prepare data and train model steps using Python script steps, an environment from a yaml file, compute targets, and pass data between steps.

Pass data between steps by creating an output file data set config, saving prep data as prep data csv, and using prep data as an input in step two.

Resubmit the pipeline from the UI or notebook to rerun jobs and monitor progress, while using the Python SDK or designer to reproduce data transformations and model training.

Publish your pipeline as an endpoint to run from any app or website, bypassing Azure login, and trigger the pipeline in Azure with a simple call.

Run the endpoint from Azure machine learning studio notebook, set up interactive login authentication and authorization header, then call the rest API of a published pipeline endpoint.

Learn to monitor a pipeline run using the run details widget in notebook and by inspecting the Azure jobs page, viewing parameters and metrics for real-time insights.

Create and deploy a model in designer by building a data pipeline, applying transformations, performing a 70/30 train-test split, and evaluating arrival delay prediction.

Learn how to remove arrival delay dependent steps in an inference pipeline, keep transformations like imputation, and deploy a model using raw input data.

Deploy a trained model to a real-time endpoint using a container instance, enable application insights and diagnostics, and validate predictions from the score dataset.

Consume a deployed model endpoint from a front-end application by sending inputs through a Python script to an API endpoint with an API key, observing real-time predictions.

Modify the inference pipeline to output only the scored labels using a Python script between the score data set and web service output, then deploy and validate the updated endpoint.

Redeploy a model by updating and republishing an endpoint to support multiple inputs while keeping the same endpoint configuration and allowing retraining with updated configurations.

Complete the DP-100 Azure data scientist associate guide by reviewing design and preparing a machine learning solution, data exploration, model training, and deployment pipelines, then schedule the Microsoft certified exam.

Explore next steps by pursuing related Microsoft certifications, such as Azure Data Engineer Associate, Azure Data Analyst Associate, and Power Platform certifications, plus DP-900.