48 Hours To Predict Pipe Failures

Machine learning competitions are ubiquitous and a great way to sharpen your skills. Many times they last for several months and have a nominal purse at stake. For anyone who has ever competed in one, they've probably found themselves spending more time then intended. There is an addictive quality in trying to eke out fractions of a percent for a target metric.

Nick Tomasino (https://www.linkedin.com/in/nicholas-tomasino-47b02622/), Piero Ferrante (https://www.linkedin.com/in/piero-ferrante-53510691/) and Matt Habiger (https://www.linkedin.com/in/matthew-habiger-5785201a/) will deep dive into their attempt to kick the addiction and crank out a solution in 48 hours to InnoCentive's Yarra Water (https://www.innocentive.com/ar/challenge/9934007) pipe prediction challenge. They'll cover how to break down a problem and define an approach on a tight timeline, divide and concur data processing tasks, how to quickly throw together a data science computing environment and generate a respectable solution. The presentation will also touch on the curse of leakage and why models that are too good to be true are usually too good to be true.