Alteryx Promote delivers AI/machine learning model deployment, management and integration

At its Inspire Europe conference in London, which kicks off today, and fresh off its acquisition of data science model management firm Yhat three months ago, Alteryx is announcing Alteryx Promote, an add-on to its Alteryx Server product, for deployment, management and integration of machine learning models, behind the corporate firewall or in the public cloud.

Also read: Alteryx expands product set, makes data science acquisition

A recurring problem
By now, the challenges of operationalizing machine learning models are starting to become better understood. Essentially, data scientists have the ability to build machine learning models and use them on a somewhat manual, bespoke basis. But deploying those models to production and making them available for mainstream developers to build intelligent applications from is another matter. For non-developers, making use of those models is even more dicey.

According to Ashley Kramer, Alteryx's VP of Product Management, Promote will address this gap by allowing deployment of models, and generation of REST APIs around them, all of which can be invoked from the Alteryx Designer environment. Kramer explained to me via email that "Models are deployed into their own customized Docker image, enabling you to customize each model based on its dependencies. Promote can then scale each model up or down based on the businesses' needs."

While Alteryx already had the ability to build, train and score against models based on the R programming language, Promote supports models built in Python, PySpark, and TensorFlow as well as R. Promote will add the ability to deploy machine learning models and generate APIs for them that are callable from a variety of application development environments.

Kramer explained that Alteryx Designer workflows can also retrain models, incorporating the latest data, then re-deploy them to Promote, adding that "Promote does a 'hot switch' [between model versions] so there is no downtime during the deployment."

Deja vu?
Other vendors have tried various approaches to address the data science operationalization issue.

Microsoft's Azure Machine Learning generates a REST-based Web service for models developed with with the cloud offering -- it even generates sample code in multiple languages that calls those APIs to run predictive queries.

Microsoft's SQL Server 2016 allows for R-based machine learning models to be stored in its databases and R code that uses the models to be hosted in its stored procedures and SQL scripts. SQL Server 2017 will add similar capabilities for Python-based models and code

My colleague Tony Baer covered recently-introduced functionality in Datameer for deploying and scoring trained models, based on Google's open source TensorFlow deep learning platform, directly from Datameer's spreadsheet big data analytics environment. (Disclosure: Datameer is my former employer and current client.)

Also read: Datameer makes deep learning more accessible

And as I reported about two weeks ago, Anaconda, which brings to market what is arguably the leading Python distribution, recently announced model operationalization capabilities of its own.

Also read: AI to the people: Anaconda readies its Enterprise 5 Python distribution

Let's all get along
In each case, the goal is the same: take machine learning models developed using otherwise esoteric languages, libraries and tools and deploy them so they're usable by mainstream developers and business users who lack data science skill sets. Each product has done this by making machine learning models work within the context of its own functionality.

Alteryx's approach, aided by its Yhat acquisition, conforms to the others, and it further validates them. With them, it is helping move AI, data science and machine learning out of the laboratory and into the realm of DevOps and self-service analytics. And there's nothing but upside to that prospect.