Water system workers with kdb+ historical database

Kdb+ Use Case: Machine Learning Water System Maintenance Application

6 Dec 2017 | , ,
Share on:

Kdb+ is being used much more widely in machine learning applications today. Its ability to quickly ingest and process data, particularly large, fragmented datasets, is one way that developers are adding kdb+ to their technology stack of artificial intelligence and machine learning tools.

For Australian kdb+ developer Sherief Khorshid, who also develops machine learning systems, incorporating kdb+ into a predictive maintenance application gave him the edge in a hackathon win that landed him a cash prize and a contract with the Water Corporation of Western Australia.


The Ministry of Data in Perth is on a mission to promote an innovation ecosystem in Western Australia. To that end, it has started sponsoring hackathons where the brightest technologists and start-ups in the state take on some of the toughest problems facing Western Australian government agencies.

The first hackathon was held over a weekend in August 2017. Dozens of engineers gathered in a competition to solve problems posed by two government agencies: Main Roads, which oversees road access, and the Water Corporation, which manages water, wastewater and drainage.

A team led by Khorshid tackled the Water Corporation problem, which asked developers to come up with a way that the Water Corporation could identify events and triggers of potential problems in the water system early enough to reduce asset failure and enable timely maintenance within the network.

The challenge

The Water Corporation gave the hackers 20 years of performance history from three water plants. There were about 1,000 physical files for each plant. The data consisted of over 300 columns and 1.5 million rows. Over the years, the format of the data had changed several times, so it was extremely messy and inconsistent. Although the data was technically considered  “computer output,” it was not really computerized in the sense that it was not in a single format that could be simply loaded into a database and analyzed as is.

The solution

Up against a ticking clock, Khorshid chose to use kdb+ rather than Python to ingest and pre-process the data because he knew he could save himself many hours in the process. He quickly saw that he would need to use kdb+ bulk pattern matching to extract the data into consistent data representations in CSV format ready to re-ingest and analyze.

Khorshid’s team used a number of neural network and XGBoost regression models to make real-time forecasts of asset failure.  His team also created a web based GUI connecting directly to a  kdb+ instance, allowing for two-way communication and data transfer.

The application that Khorshid built had a kdb+ process powering the back-end, “pumping out” the historical water network data. The back-end pushed out one hours’ worth of data every second. In the final presentation at the hackathon, Khorshid was able to show that multiple clients from around the world were able to simultaneously run the app, and query the data.

The results

Khorshid, who originally learned kdb+ as a developer at an investment bank, found that kdb+ was the best tool for pre-processing the water plant data. “When doing a lot of data munging, it is a lot easier in kdb+ than in Python,” Khorshid said.

The value of converting previously inaccessible data into usable datasets can be a significant business benefit for organizations like the Water Corporation. The agency has been collecting SCADA data for two decades from thousands of pumps and pipes, and then simply storing the data away in an historical database that they have never been able to utilize before.

Once cleansed and normalized, that data can be incorporated into improved applications for detecting leaks; measuring the effectiveness of the water system; for demand forecasting, or for preventative maintenance, all of which can meaningfully improve water operations and their service to their customers.


Classification using K-Nearest Neighbors in kdb+

21 Jun 2018 | , , , , , ,

As part of Kx25, the international kdb+ user conference held May 18th in New York City, a series of seven JuypterQ notebooks were released and are now available on https://code.kx.com/q/ml/. Each notebook demonstrates how to implement a different machine learning technique in kdb+, primarily using embedPy, to solve all kinds of machine learning problems, from feature extraction to fitting and testing a model.

ML and credit card fraud

Machine Learning and AI: The Future Possibilities in Business

20 Jun 2018 | , , ,

At Kx25, the international kdb+ user conference held in New York City on May 18th, Alan Rozet, from Kx partner Brainpool.ai, gave a presentation about how machine learning (ML) is being used in business today. Alan is a Principal Machine Learning Engineer at Capital One with wide ranging experience in ML including in the financial and health care and consumer sectors.

Machine Learning with kdb+ blog

Dimensionality Reduction in kdb+

14 Jun 2018 | , , , , ,

Dimensionality reduction methods have been the focus of much interest within the statistics and machine learning communities for a range of applications. These techniques have a long history as being methods for data pre-processing. Dimensionality reduction is the mapping of data to a lower dimensional space such that uninformative variance in the data is discarded. By doing this we hope to retain only that data that is meaningful to our machine learning problem. In addition, by finding a lower-dimensional representation of a dataset we hope we can improve the efficiency and accuracy of machine learning models.