Occupancy Prediction and Stationing Problem

Environment Setup

1. cd environment
2. conda create env -f environment.yml
3. conda activate py39
4. pip install -r requirements.txt

Datasets

Extract the datasets to the data folder and into their respective folders.

• apc: cleaned-wego-daily.apc.parquet
• weather: darksky_nashville_20220406.csv and weatherbit_weather_2010_2022.parquet
• gtfs: alltrips_mta_wego.parquet
• traffic: inrix data, can download separately

Code Setup

1. Merge datasets
   • notebooks/preprocessing.ipynb
   • If you want to examine raw GTFS files, see data/gtfs/raw_gtfs and you can follow this article.
2. Generate additional data
   • notebook/data_generation.ipynb: If some datasets are still missing, please contact me.
   • notebook/traffic_data.ipynb: Requires inrix data and might take a long time, i just use speed estimates i previously generated: data/traffic/triplevel_speed.pickle
3. Generate models
   • Day ahead (trip level) prediction: notebooks/day_ahead_prediction.ipynb
   • Any day ahead (trip level) prediction: notebooks/any_day_prediction.ipynb
   • Same day (stop level) prediction: notebooks/same_day_prediction.ipynb
   • Boarding/Alighting (stop level) prediction: notebooks/ons_offs_models.ipynb (not yet evaluated)
4. Dash application to demonstrate output of models and visualize historical data

More information

• See slides directory (To follow)