BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
Ease of use is part of Amazon SageMaker's allure, but developers still require important AI and data analysis skills to get started with the machine learning service.
SageMaker is a managed platform for teams to quickly and easily build, train and deploy machine learning models at scale. Amazon positions SageMaker as a way to lower the development barrier for machine learning applications, but experts caution that it isn't a cure-all.
Any company that wants to use SageMaker should have a data analyst or data scientist on staff who's experienced with languages and frameworks, such as SQL, Python, R, Jupyter and TensorFlow, to clean the data sets and get them ready for use.
Even though it abstracts the infrastructure and offers prepackaged development notebooks, as well as common algorithms and frameworks, Amazon SageMaker requires some knowledge of AI workflows, said Mike Leone, an Enterprise Strategy Group analyst.
"While there is an ability to provide access to the deep data science aspects of the platform for a data scientist, there has been an emphasis on empowering developers who are not necessarily data scientists but are comfortable with the tools and workflow," Leone said.
But SageMaker won't turn a developer into a data scientist. A developer without the appropriate data analysis experience might put as much data as they can into a model, but the prediction it yields won't have any real value.
"If you are a developer, you really need to go back and learn about the fundamentals of data analysis," said Kjell Carlsson, a Forrester Research analyst.
On the other hand, the service also requires data analysts to acquire some general AWS developer knowledge. For example, Amazon SageMaker requires navigation of S3 buckets and Redshift databases. Those who use multiple native services must also be comfortable with AWS Management Console and AWS Command Line Interface.
Business analysts, a third potential community for SageMaker, also need to learn more about data analysis and AWS.
Machine learning simplicity: A potential blessing and a curse
Amazon SageMaker does give data scientists a quick and easy way to put things into production -- a relief for those who find it impossible to get enough developer help. The service automatically spins up and configures instance types across a workflow to remove the infrastructure guesswork.
AWS also provides documentation for machine learning, as well as tutorials and access to premade notebooks. While users can choose to tune hyperparameters or bring in machine learning frameworks, SageMaker minimizes the need to do so.
That simplicity can be a blessing and a curse, Leone said. If it's too simple, some organizations won't get the flexibility they need; if it's not simple enough, it can severely limit the user base to only those that are highly skilled. So far, SageMaker appears to have covered both camps.
"That ability to be simple, as well as expose all the deep data science aspects depending on your skill level, really provides organizations with the flexibility they need as they continue down the AI path," Leone said.
SageMaker is still relatively new, so organizations will still need to consider other factors, such as how their costs will scale as they deploy more models. The service will likely evolve over time, perhaps through the addition of more automation or even an app store equivalent with prebuilt models for various use cases.
In the meantime, formal training for Amazon SageMaker or for machine leaning in general may not be the best route to get familiarized with the technology, said Torsten Volk, analyst at Enterprise Management Associates. Yes, it's important to know the basics and a class can help with that, but there's no substitute for experience.
"There's never an answer in advance unless you've done exactly the same thing before," he said.