Predict delays of Berlin public transport vehicles. Work in progress, made during Hamburg Mobility Hackathon 2017. We generate predictions in multiple stages:

1. Record departures with realtime delays, e.g. using record-vbb-delays.
2. Transform the data into a training set, which is just a list of individual data points. Each data point consists of the input vector (one value per feature to be recognized) and a desired output value. node index.js will generate an ndjson training set.
3. Split the raw training set into two disjunct sets, one for training, one for auditing the predictions made later. node split-into-sets.js will do this.
4. Train a neural network. More details to be added by @juliuste.
5. Generate predictions by applying the trained network to current realtime departures or future scheduled departures. Not done yet.

git clone https://github.com/derhuerst/predict-vbb-delays.git
cd predict-vbb-delays
npm i

You should now be able to run the scripts in this repo.

todo

If you have a question or have difficulties using predict-vbb-delays, please double-check your code and setup first. If you think you have found a bug or want to propose a feature, refer to the issues page.