DVN Hackathon Compressor failure prediciton problem
Deployed publicly at: (application) https://greidyapp.azurewebsites.net
(api) https://greidyapi.azurewebsites.net
Technologies used
- Angular
- Angular Material
- SignalR
- ASP.NET
- Azure ML studio
- Azure App Services
Things we learned
- Data Science is wicked hard
- Azure ML Studio let us forget about tooling and focus on solving the problem
- Azure ML Studio has a nice built-in REST service feature for trained models, super easy
- SignalR was new to most of the team, and websockets can make for a very nice user experience
- Data Science is full of dark magicks
What we did
- Tried many different ways to build a decent model for the provided dataset
- Did a ton of analysis and data cleaning to arrive at a model that is predictive in some cases.
- Website lets you upload one of the provided TSV files, and the service will run it through a Machine Learning model hosted in ML Studio to try and predict the likelihood of failure.
- Any user connected to the site when an upload is finished will see the results, like magic.
- A user can "acknowledge" the alert about the potentially problematic compressor, this removes it from the view for everyone.