Andrea Danti - Fotolia
Amazon Machine Learning is an AWS cloud service that offers visualization tools and wizards to help developers create models for machine learning use cases. From there, it's a relatively straightforward matter to pull predictions out of an application using simple APIs. Amazon and some of its competitors are taking a once challenging field that required special programming and math skills and making it accessible to almost anyone.
"Information analytics is on every company's to-do list in 2016," said Jason McKay, senior vice president and CTO of Logicworks, an AWS managed services partner based in New York City. "[Amazon] Machine Learning is another example of AWS taking something that is complex, expensive and usually slow, and turning it into an easily digestible service offering using tools originally developed by their own teams."
The services build on the algorithms that Amazon data scientists and the Amazon e-commerce business used internally, explained Laith Al-Saadoon, cloud solutions architect at CorpInfo, an AWS premier consulting partner based in Santa Monica, CA. These algorithms support billions of predictions daily in both real-time and batch operations.
The service has a pay-as-you-go approach, so there are no upfront licenses or commitments. You pay per request, and avoid having to make hardware or software commitments, Saadoon noted. In his estimation, the learning curve is not steep. As long as an administrator or data analyst understands the AWS ecosystem -- such as Kinesis, Redshift and Relational Database Service -- there is very little ramp up. IT teams can improve the pace of exploration and experimentation by using Amazon Machine Learning, thereby reducing time to market, he added.
Machine learning use cases and competition
Machine learning -- Amazon Machine Learning, in particular -- is a step up from traditional analytics because it focuses on predictive and potentially prescriptive results. Amazon Machine Learning is not as sophisticated as some other machine learning options. For instance, training data sets are limited in size to 100 gigabytes, and actual batch prediction data sets are limited to 1 terabyte.
Certainly there are many alternatives. Perhaps the best known is IBM Watson Analytics, which provides data visualization and predictive analytics with a simple "conversational" interface. Microsoft Azure Machine Learning Studio offers a library of ready-to-use examples that can be adapted for many machine learning use cases. Regardless, Saadoon is enthusiastic about Amazon and said the uses of the offering are only limited by one's imagination.
For example, most organizations want to have a better understanding of user behavior on their website to optimize their visits or hold users' attention. Machine learning and predictive analytics can analyze clickstream data using Simple Storage Service (S3) or Redshift and develop predictions about where users are likely to click. Based on those probabilities, admins or data analysts work to push those clicks to a preferred end state, such as heading toward the checkout.
Amazon Machine Learning also has potential to be used as a recommendation engine. "A lot of verticals can benefit from something like an app that recommends what restaurant a customer should visit," Saadoon said. In this example, a user might enter her age, gender and location, and the machine learning capability would compare the data to similar individuals, reviews from those individuals and similar data to come up with businesses that fit the profile. The process can be applied to a wide range of customer decisions.
Arne Sund, a blogger in Norway, wrote about one of many interesting Amazon Machine Learning use cases, applying it to predict the weather -- or, at least, the local temperature in the country's capital city. Noting that weather patterns in Oslo usually come from the west, he included observations from cities like Stavanger and Bergen in the data set. After implementing that historic data into Amazon Machine Learning, Sund found the service was able to take incoming fresh data and produce a generally accurate estimate of temperature likely to occur in the city.
"It has been very interesting to get started with Amazon Machine Learning and test it with a real-life data set," Sund wrote in a blog post.
For now, data for analysis needs to reside in an AWS resource, Saadoon noted. "That is one challenge with Amazon Machine Learning; it can only connect to S3 for object storage, Redshift for data warehousing and Relational Database Service," which offers a traditional SQL relational database.
However, data analysts who prefer to work with other data sources can use Amazon Elastic MapReduce in conjunction with an open source machine learning system, such as Apache Spark. "You could use that combination to connect to almost any Hadoop data source," he noted.
Setting up Amazon Machine Learning is pretty straightforward, Saadoon said. For end users, there is even the possibility of incorporating natural language queries using an API.
Amazon Kinesis plays a role in big data analytics
Digest big data with these AWS analytics tools
Cloud users are fascinated by data analytics