Warsaw Bike-Sharing Daily Periods Graph Dataset for Graph Neural Network (GNN) - Season 2023
Description
1. Graph Structure and Attributes These data are represented as directed graphs (DiGraphs) where each station in the Warsaw bike-sharing system (BSS) is a node. Node attributes include coordinates (lat, lng), average bike availability (avg_bikes, avg_freeRacks), weather conditions (temp, precipitation, wind_speed), and distances to points of interest (d_city_cen, d_metro_st, etc.). Additional binary flags (e.g., is_monday, is_holiday) indicate time- or date-based factors. Edges capture flows between stations, with attributes such as trips_count plus “_start” and “_end” fields (e.g., temp_start, temp_end) reflecting conditions at each node. 2. Data Collection Readings were taken every 5 minutes from May 6 to November 29, 2023, totaling 18.39 GB of JSON files. Each record includes station info (bike/rack counts), plus enriched data: weather (via a weather API), distances to infrastructure (via Google Maps Places API), and urban indicators (population density). Flows were aggregated to form edges. To reduce complexity, five daily intervals (morning rush, midday, afternoon peak, evening, night) were defined, yielding 1,039 graphs instead of 5,000. 3. Access to Files Each file (e.g., afternoon_peak_01_06_2023.pt) contains a single DiGraph with 295 nodes and 1684 edges. Users can download files and load them in Python: "import pickle with open("afternoon_peak_01_06_2023.pt", "rb") as f: G = pickle.load(f)" Files are provided in a NetworkX-compatible format; selected files are also adapted for PyTorch Geometric. 4. Additional Information This dataset captures real bike usage in Warsaw, useful for spatiotemporal analysis, machine learning, and urban mobility research. No personal IDs are included; only aggregated trip counts. Attributes such as d_city_cen or conditions offer context for demand forecasting and infrastructure-impact studies. If you find this dataset helpful, please cite it as indicated in the repository. License and usage terms are described in the accompanying documentation, ensuring open-access for academic and non-commercial use.