Buses

Software used in MSc Data Science Dissertation Project: "Limitations of Spatiotemporal Bus Prediction in a Town with Infrequent Buses".

Installation

This project assumes Python 3.7 and Anaconda are already installed.

environment.yml file contains the full lost of installed packages.

Usage

All code requires bus data from Trapeze Group which I am not free to share openly.

Assuming that you have independent access to this data the following sequence is recommended.

1. Use pipeline/bournemouth_input/data_reader.py To load and parse the data and convert into into stop events.
2. Use the various scripts in pipeline/feature_engineering/* to add features and derive new derivative files such as time series and correlation files. Recommended order:
   • filter_rate_and_overtakes.py
   • train_validate_test.py
   • add_features.py
   • add_geo_features.py
   • add_prev_next.py
   • add_offsets.py
   • others as needed.
3. Explore and test the various methods using Jupyter Notebooks in the Data Exploration folder. Some files of note:
   • Exploratory Data Analysis
     • IEA\Stop_events EDA2.ipynb Early investigations of stop events, mostly simple patterns.
     • IEA\just segments.ipynb Looking at segment based statistics.
     • IEA\Contour plots.ipynb Looking the wider context.
   • Spatiotemporal Models
     • short term\Correlation Coefficients.ipynb Looking at correlations for exogenous modelling.
     • short term\Correlation Coefficients2.ipynb Looking at correlations for exogenous modelling.
     • short term\first predictons.ipynb Using exogenous models.
     • short term\brute force predictors.ipynb Wrapper models.
     • GPS-speed\GPS-speed.ipynb Using GPS speed to estimate durations.
4. Run pipeline/models/Exogenous Models.py to generate the most positive results.

License

Licensed for non-commercial usage only. See Licence.md