YahooBousaiCrowd data contrains the following files:
- density_tokyo_20170401_20170709_30min.npy
- flowio_tokyo_20170401_20170709_30min.npy
- density_osaka_20170401_20170709_30min.npy
- flowio_osaka_20170401_20170709_30min.npy
- temporal_index.csv
- tokyo_spatial_index.csv
- osaka_spatial_index.csv
- readme.md
Each .npy data is a 4D tensor described as follows:
- timestep, height, width, channel = data.shape[0], data.shape[1], data.shape[2], data.shape[3]
- tokyo_density.shape, tokyo_flowio.shape = (4800, 80, 80, 1), (4800, 80, 80, 2)
- osaka_density.shape, osaka_flowio.shape = (4800, 60, 60, 1), (4800, 60, 60, 2)
- For timestep index, please check temporal_index.csv for the detail.
- For height&width index, please check tokyo_spatial_index.csv and osaka_spatial_index.csv for the detail.
- For density tensor, channel = 1, for flowio tensor, channel = 2, first channel is inflow, and second channel is outflow.
The spatiotemporal specification is summarized as follows:
- start_date = '2017-04-01 00:00:00'
- end_date = '2017-07-09 23:30:30'
- time_interval = '30min'
- tokyo_mesh: MINLat = 35.5, MAXLat = 35.82, MINLon = 139.5, MAXLon = 139.9, DLat = 0.004, DLon = 0.005
- osaka_mesh: MINLat = 34.58, MAXLat = 34.82, MINLon = 135.35, MAXLon = 135.65, DLat = 0.004, DLon = 0.005
The maximum value for the current 0-1 normalized 4D tensor is as follows:
- tokyo_density, tokyo_flowio, osaka_density, osaka_flowio = 2300, 1200, 1800, 770
- The original maximum values used in the paper is 23xx, 12xx, 18xx, 77x.
- However, according to the company policy of Yahoo Japan Corporation, only the most significant two digits can be published.
- Thus, the reproduced results might slightly differ from the paper.