Chapter 5: Automated Machine Learning with Microsoft Azure

"By far, the greatest danger of artificial intelligence is that people conclude too early that they understand it."

– Eliezer Yudkowsky

The Microsoft Azure platform and its associated toolset are diverse and part of a larger enterprise ecosystem that is a force to be reckoned with. It enables businesses to focus on what they do best by accelerating growth via improved communication, resource management, and facilitating advance actionable analytics. In the previous chapter, you were introduced to the Azure Machine Learning platform and its services. You learned how to get started with Azure machine learning, and you took a glimpse at the end-to-end machine learning life cycle using the power of the Microsoft Azure platform and its services. That was quite literally (in the non-literal sense of the word) the tip of the iceberg.

In this chapter, we will get started by looking at Automated Machine Learning...

AutoML in Microsoft Azure

AutoML is treated as a first-class citizen in the Azure platform. The fundamental ideas behind feature engineering, network architecture search, and hyperparameter tuning are the same as what we discussed in Chapter 2, Automated Machine Learning, Algorithms, and Techniques, and Chapter 3, Automated Machine Learning with Open Source Tools and Libraries. However, the layer of abstraction that's used to democratize these skills makes them much more appealing to non-machine learning experts.

The key principles of AutoML in the Azure platform are shown in the following diagram. User input such as datasets target metrics, and constraints (how long to run the job, what the allocated budget is for compute, and so on) drive the AutoML "engine", which completes iterations to find the best model and rank it according to the score of Training Success:

Figure 5.1 – Azure AutoML workflow – how AutoML works

In this...

Time series prediction using AutoML

Forecasting energy demand is a real problem in the industry where energy providers like to predict the consumer's expected needs in advance. In this example, we will use the New York City energy demand dataset, which is available in the public domain. We will use historic time series data and apply AutoML for forecasting; that is, predicting energy demand for the next 48 hours.

The machine learning notebook is part of the Azure model repository, which can be accessed on GitHub at https://github.com/Azure/MachineLearningNotebooks/. Let's get started:

1. Clone the aforementioned GitHub repository on your local disk and navigate to the forecasting-energy-demand folder:

   Figure 5.30 – Azure Machine Learning notebooks GitHub repository

2. Click on the Upload folder icon and upload the forecasting-energy-demand folder to the Azure notebook repository, as shown in the following screenshot:

   Figure 5.31 – Uploading a folder...

Summary

In this chapter, you learned how to apply AutoML in Azure to a classification problem and a time series prediction problem. You were able to build a model within the Azure Machine Learning environment with an Azure notebook and via JupyterLab. You then understood how the entire workspace relates to the experiments and runs. You also see the visualization during these automated runs; this is where feature importance, the global and local impact of features, and explanations based on raw and engineered features provide an intuitive understanding. Besides your affinity with a tool, it is also important that the platform aligns with your enterprise roadmap. Azure is an overall great platform with a comprehensive set of tools, and we hope you enjoyed exploring its automated ML capabilities.

Further reading

For more information on the topics that were covered in this chapter, please take a look at the following links:

• Azure AutoML:

  https://docs.microsoft.com/en-us/azure/machine-learning/concept-automated-ml

• Practical AutoML on Azure:

  https://github.com/PracticalAutomatedMachineLearning/Azure