Dataset Viewer
Auto-converted to Parquet Duplicate
bytes
int64
file
string
format
string
notes
list
sha256
string
source
dict
stats
dict
tables
dict
30,448,600
datasets/pek.h5
pandas HDF5 (PyTables)
[ "coordinate table key is 'poi'", "the upstream run.py reads key='coor'; use the key above instead", "coordinates are normalized, not WGS84 degrees" ]
7b09236f4dfbb0d5f52891b1cd49270296dde79fb66e0300496c556954de7b47
{ "license": "MIT", "original_file": "datasets/pek.h5", "paper": "https://doi.org/10.1609/aaai.v35i5.16548", "project": "CTLE", "repository": "https://github.com/Logan-Lin/CTLE" }
{ "coord_rows": 7274, "days": 5, "duplicate_rows": 18489, "lat_range": [ 0, 1 ], "lng_range": [ 0, 1 ], "null_values": { "datetime": 0, "poi_id": 0, "user_id": 0 }, "pois": 7274, "records": 944763, "records_per_day": { "2019-09-09": 178194, "2019-09-10": 19646...
{ "data": { "columns": [ "user_id", "poi_id", "datetime" ], "dtypes": { "datetime": "datetime64[ns]", "poi_id": "int64", "user_id": "int64" }, "rows": 944763 }, "poi": { "columns": [ "lat", "lng" ], "dtypes": { "lat": "float64",...

CTLE-PEK

The pek example trajectory dataset shipped with CTLE: Context and Time-aware Location Embedding, from the AAAI 2021 paper Pre-training Context and Time Aware Location Embeddings from Spatial-Temporal Trajectories for User Next Location Prediction.

This repository mirrors the original datasets/pek.h5 unchanged. The file is pandas/PyTables HDF5 with two tables:

Key Rows Columns Meaning
/data 944,763 user_id, poi_id, datetime mobile-signaling / trajectory records
/poi 7,274 lat, lng (index: poi_id) location coordinates

Statistics

Item Value
Records 944,763
Users 8,319
POIs 7,274
Days 5
Time range 2019-09-09T04:30:00 — 2019-09-13T22:30:00
Latitude range 0.0 — 1.0
Longitude range 0.0 — 1.0

Coordinates in this example file are normalized to [0, 1], not raw WGS84 degrees. The file also contains 18,489 exact duplicate rows, preserved from the original release.

Usage

Download the file (HuggingFace):

from huggingface_hub import hf_hub_download
import pandas as pd

path = hf_hub_download("Davidham3/ctle-pek", "datasets/pek.h5", repo_type="dataset")
raw = pd.read_hdf(path, key="data")
poi = pd.read_hdf(path, key="poi")

ModelScope:

from modelscope import dataset_snapshot_download
import pandas as pd

root = dataset_snapshot_download("Davidham3/ctle-pek")
raw = pd.read_hdf(f"{root}/datasets/pek.h5", key="data")
poi = pd.read_hdf(f"{root}/datasets/pek.h5", key="poi")

Or download the whole dataset into an existing CTLE checkout:

hf download Davidham3/ctle-pek --repo-type dataset --local-dir CTLE-main

Important: key name

The upstream run.py currently reads the coordinate table with key="coor", but the shipped pek.h5 actually stores it under key="poi". Use key="poi", or rename the key, when running the original code against this file.

Provenance

Downloads last month
54