Warsaw Bike-Sharing Daily Periods Graph Dataset for Graph Neural Network (GNN) - Season 2023

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.

1. 图结构与属性
本数据集以有向图（DiGraphs）形式表示，华沙自行车共享系统（BSS）中的每个站点对应一个节点。节点属性包含坐标（纬度lat、经度lng）、平均单车可用量（avg_bikes）与空闲车位数（avg_freeRacks）、天气状况（气温temp、降水量precipitation、风速wind_speed），以及至各类兴趣点的距离（d_city_cen、d_metro_st等）。此外还包含若干二元标志位（例如is_monday、is_holiday），用于表征时间或日期相关的影响因素。图边用于捕捉站点间的流量，其属性包含出行总量（trips_count）以及以“_start”和“_end”结尾的字段（如temp_start、temp_end），分别反映对应节点处的环境条件。

2. 数据采集
数据采集周期为2023年5月6日至11月29日，每5分钟采集一次，总数据量达18.39 GB，存储为JSON格式文件。每条记录包含站点信息（单车与车位数统计），以及丰富的补充数据：天气数据（通过气象API获取）、至城市基础设施的距离（通过谷歌地图地点API（Google Maps Places API）获取），以及城市表征指标（如人口密度）。站点间的流量经聚合后形成图边。为降低计算复杂度，我们将每日划分为五个时段：早高峰、午间、晚高峰、晚间以及夜间，最终将原始的5000余份数据聚合为1039张有向图。

3. 文件获取与加载
单个文件（例如afternoon_peak_01_06_2023.pt）仅包含一张有向图（DiGraphs），该图包含295个节点与1684条边。用户可下载文件并在Python环境中加载，示例代码如下：
"import pickle

with open("afternoon_peak_01_06_2023.pt", "rb") as f:
G = pickle.load(f)"
所有文件均采用兼容NetworkX的格式存储，部分精选文件同时适配PyTorch Geometric框架。

4. 补充说明
本数据集记录了华沙市真实的单车使用数据，可用于时空分析、机器学习以及城市出行研究等场景。数据中未包含任何个人身份信息，仅保留聚合后的出行总量。诸如d_city_cen、环境条件等属性可为出行需求预测以及基础设施影响评估研究提供上下文支撑。若您在研究中使用本数据集，请按照仓库中说明的方式进行引用。数据集的许可协议与使用条款详见配套文档，允许学术与非商业用途的开放获取。