Azure Machine Learning Studio
Azure Machine learning Studio is a cloud based integrated environment for developing Machine Learning solutions which combines no-code/low code and code first experiences for an inclusive data scientist platform.
Let me explain bit more about ML Studio (Classic) and Azure Machine Learning Studio.
ML Studio (Classic)
By url https://studio.azureml.net/, was released in 2015 and was first drag and drop machine learning builder.
- Can manage datasets
- Can create ML model with drag and drop with low code options
- Can add R and Python code module
- Can manage trained model
- Publish model as web services
With ML Classic, we can design ML model with drag and drop interface and then publish it from same platform. This studio supports experiment with scalable up to 10GB training data limit only. In terms of deployment computing, ML classic studio Solutions can be published in web service format and are not customizable. Considering MLOps, this studio can do basic model management and CPU deployments only. The below features are not supported in ML Studio (Classic):
- Code SDKs
- ML Pipeline
- Automated Model training and hyperparameter tuning
- Data Drift detection
- Data labeling projects.
I will depict a complete ML solution using this studio in next article.
Azure machine Learning Studio
This, by url https://ml.azure.com/, is a modern studio which outperforms over the classic studio and has several exclusive features. Therefore, this studio is a complete package for developing and deploying ML solutions of any nature. The platform can handle all scenarios of Machine learning, starting from Classic ML to deep learning, and is also suitable for supervised as well as an unsupervised approach. You can go for code first approach with the full flavor of writing Python or R code with the SDK. You have the option to create a workspace with notebooks (Jupiter Notebooks), taking full advantage of Python code, likewise for RStudio.
The studio gives wide range of options to manage your assets and resources directly from browser with seamless experience. Using this studio, we can manage:
- Computer resources
- Run logs
- Pipeline Endpoints.
Notebooks is available for author machine learning projects to write and run your own code in managed Jupyter Notebook servers that are directly integrated in the studio. Using python SDKs, we can write, run and take full advantages python code.
Additionally, Azure Machine Learning designer is included for no code or low code, drag and drop to create ML pipelines.
Another exclusive option, AutoML comes with this studio. Let me explain a bit more about AutoML.
Automated machine learning, also referred to as automated ML or AutoML, is the process of automating the time consuming, iterative tasks of machine learning model development. It allows data scientists, analysts, and developers to build ML models with high scale, efficiency, and productivity all whilesustaining model quality. Automated ML is based on a breakthrough from our Microsoft Research division.
Traditional machine learning model development is resource-intensive, requiring significant domain knowledge and time to produce and compare dozens of models. With automated machine learning, you’ll accelerate the time it takes to get production-ready ML models with great ease and efficiency.
However, we can use Automate ML for all types of ML projects. Auto ML can be used for following ML type:
- Time-series forecasting
How AutoML Works.
Model creation is typically time consuming and iterative process. Most of the cases, the desired accuracy cannot be achieved with one model.
We choose some features from the data, and select the algorithm then set parameters, finally run the model to check the score or evaluation (accuracy). If the accuracy is below the expectation then we change the features, choose another algorithm or change the parameters, and again run the model to check score. This iterative process will continue several times until we get desired accuracy which is time consuming.
However, automated machine learning can accelerate the model development by automating the iterative tacks. You can prove dataset, set target metric and constraints, AutoML will select different features, algorithms and parameters run the experiments. The process will automatically repeat with different features, algorithms and parameters and at last, it will list the results with score and will show top 3 model as depicted.
Data labeling is another authoring option which is available to efficiently coordinate data labeling projects with the use of Azure Machine Learning data labeling.
Deploy your existing model with Azure machine learning
This option is basically for deployment as cloud service. If you have your ML model trained outside of Azure Machine learning, you can register and deploy the model in azure as web service or to an IoT Edge device. After deployment, you can monitor your model and detect data drift in Azure ML.
MLOps (Machine Learning Operations)
Machine Learning Operations (MLOps) is based on DevOps principles and practices that increase the efficiency of workflows. For example, continuous integration, delivery, and deployment. MLOps applies these principles to the machine learning process, with the goal of:
- Faster experimentation and development of models
- Faster deployment of models into production
- Quality assurance
ML Studio (classic) is Azure ML first drag-and-drop machine learning builder, however, it does not interoperate with Azure Machine Learning. On the other hand, Azure Machine Learning is a separate complete platform for data science with modernized services and several features. Azure Machine Learning Studio is a web-based portal in Azure Machine Learning for developing and managing options of ML model for project authoring and asset management.