I want to start by saying that I am not an expert in Machine Learning or Cloud. I used the documentation from the Microsoft website to help me set up this example.
The first step was to go to portal.azure.com and log in. I created a resource group called “Machine Learning” and set my budget alerts. The search bar on the Azure portal is conveniently always at the top, so I searched for “Azure Machine Learning”. Here, I created a workspace called “Healthcare2”. I needed a simple dataset to train, so I chose the Diabetes.csv data from Kaggle and added it to my workspace.
After I added my data, the schema form allowed me to remove or edit the columns pulled from the CSV. The dataset was then processed, and the Explore tab showed that the data was uploaded correctly.
The next step was to use Automated Machine Learning. AutoML uses the data to create multiple models in minutes and provides the weighted score. From the home screen, I clicked on the Automated ML option. I clicked the “New Automated ML job” and chose a Regression task to predict future values. I picked the healthcare dataset from the list, then I picked a target column with integers to use. A compute resource was added to run commands to train the model. I picked the least expensive one since my dataset was small. When I submitted the training job, it started the compute resource, built the machine learning pipeline, and created the models.
Now that the model has been created, I have checked the Models and Output + Logs tab. The output provides the code used to make the pipeline; if there were any logs, they would contain the errors. This is a great way to tune a model for a specific metric and easily identify problems within the ML pipeline.
In Conclusion
The Azure Machine Learning Studio does a lot of complex tasks that could cut down a lot of development time for an engineering team. It does the data scientist's work of cleaning and interpreting the data and building the pipeline for automation like a software engineer. It's built for collaborative teams requiring minimal legwork and looking to implement quickly.