My "project that doesn't make me money" is https://transitfeeds.com
Currently it archives a ton of static schedule data, and some basic GTFS-realtime archiving stuff.
My longer-term goal is to archive all the GTFS-rt feeds every 30 seconds to provide similar analysis to what you've done here.
Obviously this requires even more storage than I'm already using and a ton of data processing, which is probably above my pay grade.
ty
Super easy automated machine learning in Python. Fast enough to run at production speeds. I'm the author, and would love any feedback you have! The whole point of the project is to make ML available to people like you who just like building things, and think their thing could be better with a bit of ML.
I made this septa app years ago when realtime-ish data became available.. http://septa.kaybox.org/
Displays all on-time trains as blue, and late trains red. Locations update every 30 seconds, I think.
This lets public transit passengers answer questions like:
- "My train is getting later and later, is it actually moving?"
- "My train is getting later and later, has it actually STARTED its journey?" (sometimes the answer is "no", sadly)
- "Is it just my train, or are many trains running late?"
- "What was the on-time performance of this train like yesterday? 2 days ago? 7 days ago?" (Some trains tend to be chronically late)
It may come as a surprise that the backend of the system is actually not a database, but Splunk (http://www.splunk.com). DBs are nice, but Splunk is fantastic when it comes to data analytics and reporting.
I'm currently waiting for Splunk to make some of their machine learning modules available for free so that I can start pulling in weather data, train the machine learning component against both that and the train data, and use that to predict the likelihood of any given train becoming late.