Datameer makes deep learning more accessible

With SmartAI, Datameer is addressing the last mile in putting machine learning to work in business intelligence.
Written by Tony Baer (dbInsight), Contributor

Video: How Haven Life uses AI, machine learning to spin new life out of long-tail data

Given the spotlight on machine learning and AI, it's natural to ask the question, "Now what?" The challenge has been how to take AI models from the laptop to production and deliver business value.

As nature abhors a vacuum, there's been no shortage of solutions for addressing pieces of the gap. IBM's Watson Data Platform is an ambitious framework of solutions addressing the lifecycle from collaboration to operationalization. Cloudera's Data Science Workbench aims to move experimentation with algorithms from laptop to the Hadoop cluster. Data science collaboration tools from providers like Dataiku, Domino Data Lab, and Alpine Data target collaboration, workflow, and the lifecycle management of data science and machine learning models. Meanwhile, Alteryx lets you embed R programs under the hood of a self-service BI tool.

So there's no silver bullet to bridging the gap from the mind of the data scientist to the everyday incorporation of models into day to day operational analytics. With SmartAI, Datameer is addressing the last mile. It adds the capability to "bring your own model" into Datameer and run it as a Datameer analytical spreadsheet function. Specifically, SmartAI imports binaries for models developed using TensorFlow, the deep learning library that has been open sourced by Google.

In practice, that means that once your data scientist or data science team has tested and validated models, it can then be dropped into a Datameer analytic pipeline. And that's where the analytic lifecycle kicks in, beginning upstream with data preparation, integration, and feature engineering, and then executing the model by invoking it as a Datameer analytic function (the tool has a library of over a couple hundred functions). So a deep learning model can be applied to specific business problems, such as Customer 360, genomic data analysis, operational monitoring, or fraud detection. Closing the loop, the data sets can be used for training and refining the models.

So Datameer will provide the straightest path from developing a TensorFlow deep learning model to embedding it into a BI application. But that prompts the next question: how will customers take advantage of it?

If you drew a heat map around chatter and buzz, machine learning is front and center. But machine learning is a vast umbrella of approaches, spanning from the intelligent pattern matching of clustering, random forest, or path analysis approaches, to the more ambitious approaches of deep learning and cognitive computing that border on human perception and thought processes.


By targeting TensorFlow, Datameer has chosen the library that has drawn significant interest from developers. It's a shrewd strategy, especially if Datameer were seeking to differentiate itself on the emerging Google Cloud platform. But given that deep learning is territory not as well charted compared to less ambitious machine learning approaches, we wonder how much practical advantage Datameer customers will realize in the short run.

Nonetheless, by enabling a form of plug and play to machine learning models, Datameer is clearing a pathway for making the benefits tangible through a BI tool that's within the comfort zone of business analysts. The good news is that the connector could support similar integration to other popular machine learning libraries as well. If Datameer does so with Spark MLlib or others, it would open the floodgates to machine learning BI applications a lot wider.

Editorial standards