Migrate to Azure Machine Learning from Studio (classic)

Important

Support for Machine Learning Studio (classic) ends on 31 August 2024. We recommend that you transition to Azure Machine Learning by that date.

After December 2021, you can no longer create new Studio (classic) resources. Through 31 August 2024, you can continue to use existing Studio (classic) resources.

Studio (classic) documentation is being retired and might not be updated in the future.

Learn how to migrate from Machine Learning Studio (classic) to Azure Machine Learning. Azure Machine Learning provides a modernized data science platform that combines no-code and code-first approaches.

This guide walks through a basic lift and shift migration. If you want to optimize an existing machine learning workflow, or modernize a machine learning platform, see the Azure Machine Learning Adoption Framework for more resources, including digital survey tools, worksheets, and planning templates.

Please work with your cloud solution architect on the migration.

Recommended approach

To migrate to Azure Machine Learning, we recommend the following approach:

• Step 1: Assess Azure Machine Learning
• Step 2: Define a strategy and plan
• Step 3: Rebuild experiments and web services
• Step 4: Integrate client apps
• Step 5: Clean up Studio (classic) assets
• Step 6: Review and expand scenarios

Step 1: Assess Azure Machine Learning

1. Learn about Azure Machine Learning and its benefits, costs, and architecture.

2. Compare the capabilities of Azure Machine Learning and Studio (classic).

   The following table summarizes the key differences.

   Feature	Studio (classic)	Azure Machine Learning
   Drag-and-drop interface	Classic experience	Updated experience: Azure Machine Learning designer
   Code SDKs	Not supported	Fully integrated with Azure Machine Learning Python and R SDKs
   Experiment	Scalable (10-GB training data limit)	Scale with compute target
   Training compute targets	Proprietary compute target, CPU support only	Wide range of customizable training compute targets; includes GPU and CPU support
   Deployment compute targets	Proprietary web service format, not customizable	Wide range of customizable deployment compute targets; includes GPU and CPU support
   Machine learning pipeline	Not supported	Build flexible, modular pipelines to automate workflows
   MLOps	Basic model management and deployment; CPU-only deployments	Entity versioning (model, data, workflows), workflow automation, integration with CICD tooling, CPU and GPU deployments, and more
   Model format	Proprietary format, Studio (classic) only	Multiple supported formats depending on training job type
   Automated model training and hyperparameter tuning	Not supported	Supported Code-first and no-code options
   Data drift detection	Not supported	Supported
   Data labeling projects	Not supported	Supported
   Role-based access control (RBAC)	Only contributor and owner role	Flexible role definition and RBAC control
   AI Gallery	Supported	Not supported Learn with sample Python SDK notebooks

   Note

   The designer feature in Azure Machine Learning provides a drag-and-drop experience that's similar to Studio (classic). However, Azure Machine Learning also provides robust code-first workflows as an alternative. This migration series focuses on the designer, since it's most similar to the Studio (classic) experience.

3. Verify that your critical Studio (classic) modules are supported in Azure Machine Learning designer. For more information, see the Studio (classic) and designer component-mapping table.

4. Create an Azure Machine Learning workspace.

Step 2: Define a strategy and plan

1. Define business justifications and expected outcomes.

2. Align an actionable Azure Machine Learning adoption plan to business outcomes.

3. Prepare people, processes, and environments for change.

Please work with your cloud solution architect to define your strategy.

For planning resources, including a planning doc template, see the Azure Machine Learning Adoption Framework.

Step 3: Rebuild your first model

After you define a strategy, migrate your first model.

1. Migrate a dataset to Azure Machine Learning.

2. Use the Azure Machine Learning designer to rebuild an experiment.

3. Use the Azure Machine Learning designer to redeploy a web service.

   Note

   This guidance is built on top of Azure Machine Learning v1 concepts and features. Azure Machine Learning has CLI v2 and Python SDK v2. We suggest that you rebuild your Studio (classic) models using v2 instead of v1. Start with Azure Machine Learning v2.

Step 4: Integrate client apps

Modify client applications that invoke Studio (classic) web services to use your new Azure Machine Learning endpoints.

Step 5: Clean up Studio (classic) assets

To avoid extra charges, clean up Studio (classic) assets. You might want to retain assets for fallback until you've validated Azure Machine Learning workloads.

Step 6: Review and expand scenarios

1. Review the model migration for best practices and validate workloads.

2. Expand scenarios and migrate more workloads to Azure Machine Learning.

Studio (classic) and designer component-mapping

Consult the following table to see which modules to use while rebuilding Studio (classic) experiments in the Azure Machine Learning designer.

Important

The designer implements modules through open-source Python packages rather than C# packages like Studio (classic). Because of this difference, the output of designer components might vary slightly from their Studio (classic) counterparts.

Category	Studio (classic) module	Replacement designer component
Data input and output	- Enter data manually - Export data - Import data - Load trained model - Unpack zipped datasets	- Enter data manually - Export data - Import data
Data format conversions	- Convert to CSV - Convert to dataset - Convert to ARFF - Convert to SVMLight - Convert to TSV	- Convert to CSV - Convert to dataset
Data transformation – Manipulation	- Add columns - Add rows - Apply SQL transformation - Clean missing data - Convert to indicator values - Edit metadata - Join data - Remove duplicate rows - Select columns in dataset - Select columns transform - SMOTE - Group categorical values	- Add columns - Add rows - Apply SQL transformation - Clean missing data - Convert to indicator values - Edit metadata - Join data - Remove duplicate rows - Select columns in dataset - Select columns transform - SMOTE
Data transformation – Scale and reduce	- Clip values - Group data into bins - Normalize data - Principal component analysis	- Clip values - Group data into bins - Normalize data
Data transformation – Sample and split	- Partition and sample - Split data	- Partition and sample - Split data
Data transformation – Filter	- Apply filter - FIR filter - IIR filter - Median filter - Moving average filter - Threshold filter - User-defined filter
Data transformation – Learning with counts	- Build counting transform - Export count table - Import count table - Merge count transform - Modify count table parameters
Feature selection	- Filter-based feature selection - Fisher linear discriminant analysis - Permutation feature importance	- Filter-based feature selection - Permutation feature importance
Model – Classification	- Multiclass decision forest - Multiclass decision jungle - Multiclass logistic regression - Multiclass neural network - One-vs-all multiclass - Two-class averaged perceptron - Two-class Bayes point machine - Two-class boosted decision tree - Two-class decision forest - Two-class decision jungle - Two-class locally deep SVM - Two-class logistic regression - Two-class neural network - Two-class support vector machine	- Multiclass decision forest - Multiclass boost decision tree - Multiclass logistic regression - Multiclass neural network - One-vs-all multiclass - Two-class averaged perceptron - Two-class boosted decision tree - Two-class decision forest - Two-class logistic regression - Two-class neural network - Two-class support vector machine
Model – Clustering	- K-means clustering	- K-means clustering
Model – Regression	- Bayesian linear regression - Boosted decision tree regression - Decision forest regression - Fast forest quantile regression - Linear regression - Neural network regression - Ordinal regression - Poisson regression	- Boosted decision tree regression - Decision forest regression - Fast forest quantile regression - Linear regression - Neural network regression - Poisson regression
Model – Anomaly detection	- One-class SVM - PCA-based anomaly detection	- PCA-based anomaly detection
Machine Learning – Evaluate	- Cross-validate model - Evaluate model - Evaluate recommender	- Cross-validate model - Evaluate model - Evaluate recommender
Machine Learning – Train	- Sweep clustering - Train anomaly detection model - Train clustering model - Train matchbox recommender - Train model - Tune model hyperparameters	- Train anomaly detection model - Train clustering model - Train model - Train PyTorch model - Train SVD recommender - Train wide and deep recommender - Tune model hyperparameters
Machine Learning – Score	- Apply transformation - Assign data to clusters - Score matchbox recommender - Score model	- Apply transformation - Assign data to clusters - Score image model - Score model - Score SVD recommender - Score wide and deep recommender
OpenCV library modules	- Import images - Pre-trained cascade image classification
Python language modules	- Execute Python script	- Execute Python script - Create Python model
R language modules	- Execute R script - Create R model	- Execute R script
Statistical functions	- Apply math operation - Compute elementary statistics - Compute linear correlation - Evaluate probability function - Replace discrete values - Summarize data - Test hypothesis using t-Test	- Apply math operation - Summarize data
Text analytics	- Detect languages - Extract key phrases from text - Extract N-gram features from text - Feature hashing - Latent dirichlet allocation - Named entity recognition - Preprocess text - Score vVowpal Wabbit version 7-10 model - Score Vowpal Wabbit version 8 model - Train Vowpal Wabbit version 7-10 model - Train Vowpal Wabbit version 8 model	- Convert Word to vector - Extract N-gram features from text - Feature hashing - Latent dirichlet allocation - Preprocess text - Score Vowpal Wabbit model - Train Vowpal Wabbit model
Time series	- Time series anomaly detection
Web service	- Input - Output	- Input - Output
Computer vision		- Apply image transformation - Convert to image directory - Init image transformation - Split image directory - DenseNet image classification - ResNet image classification

For more information on how to use individual designer components, see the Algorithm & component reference.

What if a designer component is missing?

Azure Machine Learning designer contains the most popular modules from Studio (classic). It also includes new modules that take advantage of the latest machine learning techniques.

If your migration is blocked due to missing modules in the designer, contact us by creating a support ticket.

Example migration

The following migration example highlights some of the differences between Studio (classic) and Azure Machine Learning.

Datasets

In Studio (classic), datasets were saved in your workspace and could only be used by Studio (classic).

In Azure Machine Learning, datasets are registered to the workspace and can be used across all of Azure Machine Learning. For more information on the benefits of Azure Machine Learning datasets, see Data in Azure Machine Learning.

Pipeline

In Studio (classic), experiments contained the processing logic for your work. You created experiments with drag-and-drop modules.

In Azure Machine Learning, pipelines contain the processing logic for your work. You can create pipelines with either drag-and-drop modules or by writing code.

Web service endpoints

Studio (classic) used REQUEST/RESPOND API for real-time prediction and BATCH EXECUTION API for batch prediction or retraining.

Azure Machine Learning uses real-time endpoints (managed endpoints) for real-time prediction and pipeline endpoints for batch prediction or retraining.

Related content

In this article, you learned the high-level requirements for migrating to Azure Machine Learning. For detailed steps, see the other articles in the Machine Learning Studio (classic) migration series:

• Migrate a Studio (classic) dataset
• Rebuild a Studio (classic) experiment
• Rebuild a Studio (classic) web service
• Consume pipeline endpoints from client applications.
• Migrate Execute R Script modules

For more migration resources, see the Azure Machine Learning Adoption Framework.