Machine learning provides businesses with the ability to develop predictive analytics for data sets. The benefits to this can range from improved internal efficiency to a more precise understanding of what their customers might do or want. Machine learning has quickly become a cutting edge tool in an enterprise's big data toolbox. A great example can be found in a white paper published by Google about how machine learning could be used to optimize data center efficiency and cut cost. Companies are realizing the many potential benefits machine learning can provide however, the cost and expertise required to make use of it can be a roadblock for some.
Azure Machine Learning
As with any cloud service the obvious benefits to be seen are from a cost and maintenance perspective. Azure Machine Learning is no exception to this, with a subscription cost of around $120 per year per seat in addition to the on demand cost it remains well below the cost incurred by other enterprise grade machine learning licenses. The on demand costs are minimal with each hour in the machine learning studio costing $1. In addition, there is no need to procure or maintain hardware for a license that a business may not use regularly. With Azure Machine Learning infrastructure is an afterthought for the user, all required hardware is ready and scaled as needed unbeknownst to the user. Azure's free tier also gives enterprises a great opportunity to experiment with Machine Learning without being charged so that they can better understand how it can be leveraged to benefit their business.
Azure Machine Learning offers other key benefits, its drag and drop interface makes it easy for users to easily build and utilize machine learning experiments with no code required. However, if a user would like to write something custom, Azure Machine Learning also employs powerful integration with R and python so existing or new scripts can be dropped in seamlessly into any part of the experiment. Another added benefit of Azure Machine Learning is the ability for a user to make use of algorithms used by Microsoft's Xbox and Bing products. The wide variety of tools provides users of all levels an innovative platform to use.
Users can drop different data inputs, outputs, converters, and transformers into the experiment; each has a range of options to help the user preprocess their data and store the results as desired. The user can pull data in from a range of resources including Azure storage containers, a data feed provider, Hive query, or over HTTP. The results can also be stored using most of the same methods.
The user can easily view the progress of an experiment as it runs through each step. At each point in the process the user can view the data and results produced. The user can then easily adjust different settings and options and quickly create several iterations of an experiment as they tinker with it. All of this is viewable in the run history of experiment so that the user can keep up with the different versions they have tried.
In addition, once experiments are built and run they can be published as web services. The user has granular control over what fields and options in the experiments are inputs for the web service; the user can give the consumer of the API the ability to build and train models before inputting their own data for analysis. Once published the API also comes with pre-made documentation for end users. Users also have the ability to sell their web services on Azure's marketplace.
Azure's marketplace already has a number of APIs for free or that can be purchased; they provide an excellent example of how machine learning can benefit any enterprise. For example, one API currently can be trained to return recommended products for customers to purchase based on historical sales data and current product catalogue. Another can be trained to predict customer churn for a business. These APIs just scratch the surface of the many possibilities that machine learning provides.
Azure Machine Learning provides enterprises with scalable, on-demand machine learning capabilities that is cost effective, maintenance free, and simple to use. The simple interface provides an easy way to create and adjust experiments quickly as well as easily visualize the data flow through the experiment. Integrating with R and python scripts makes the platform highly extensible and customizable to a user's needs; it also makes it easier for an enterprise to transition their current machine learning setup to the cloud. In aggregate, these benefits provide a compelling reason for companies to look at Azure to learn about opportunities in machine learning as well as expand and grow their current use.