A fast-growing fast food establishment needed to expand its machine learning capabilities to optimize its US-wide operations. They needed guidance to determine if using Amazon Redshift’s new automated machine learning would benefit their business, with several upcoming algorithms.
phData exposed the strengths and limitations of Amazon Redshift Machine Learning for the customer’s specific use case, recreating their algorithm within Redshift. We also identified which use cases are most appropriate to be used within the tool.
Amazon recently released a new capability in their Redshift clusters to handle automated machine learning. Amazon Redshift ML makes it easy for data analysts and database developers to create, train, and apply machine learning models using familiar SQL commands in Amazon Redshift data warehouses.
The customer already had several machine learning algorithms running, and they were planning out several more. They needed to know if running these new machine learning algorithms on Amazon Redshift’s automated tool would optimize their models (and, in turn, optimize their operations).
The proof is in the pudding, as they say. phData came on board to trial the new AWS Redshift capabilities with one of the customer’s existing models to determine if the feature would be beneficial to their business.
The customer had worked with phData on several machine learning projects in the past. They needed a quick assessment from experts they trust, so they turned to our team. (We should note: If you’re looking for help with your machine learning capabilities and tools, this kind of assessment is a perfect entry point.)
First, the technical bit: The phData team used Amazon Redshift Machine Learning, Amazon Redshift Spectrum as the integration point to load data into Redshift, and AWS Glue for creating the data catalog.
We originally planned to test the capabilities of Bring Your Own Model (BYOM) using CatBoost to recreate the Operator Led Delivery forecast model. But, upon further investigation, we discovered that BYOM is limited to inference only. This makes it impossible to train CatBoost models in Redshift ML.
As an alternative, we recreated the customer’s existing AAR model and pipeline in Amazon Redshift ML to compare directly with their existing model and pipeline.
We then brought our findings to the customer, along with the code to reproduce the machine learning model at their convenience.
phData exposed the strengths and limitations of the new Amazon Redshift ML tool. We also identified which use cases are most appropriate to be used within the new features (and which are not).
The customer was extremely pleased with the outcome—especially because it provided them with the opportunity to investigate the use of the tool without having to resource the project within their own organization.
Looking into optimizing machine learning for your organization? Learn how phData can help solve your most challenging problems.
Data Coach is our premium analytics training program with one-on-one coaching from renowned experts.