t22000t/bike-sharing-tabular

Bike Sharing Demand - Hourly (Poisson)

A ready-to-use copy of the UCI Bike Sharing Dataset (hourly granularity, 17,379 × 17), accompanied by baseline metrics from an 8-architecture tabular modelling pipeline for direct comparison.

Originally collected and published by Fanaee-T & Gama (2014). Source: UCI ML Repository id 275.

At a glance

Field	Value
Rows	17,379 hourly observations
Time range	Jan 2011 - Dec 2012
Columns	17 (16 features + 1 target)
Target	cnt (hourly bike rental count)
Target range	1 - 977
Target mean / median	189 / 142
Distribution family	Poisson (count data)
Continuous features	7 (temperature, humidity, wind, time features)
Categorical features	5 (season, weather, day-of-week, holiday, working day)
Missing values	none

Suggested distribution family

cnt is a non-negative count, so a Poisson family with log link is the natural choice. The tabular-data-modelling-pipeline ships a ready-made config: configs/example_bike_sharing.py.

How to use

from datasets import load_dataset
ds = load_dataset("t22000t/bike-sharing-tabular", split="train")
print(ds[0])

Or plain pandas:

import pandas as pd
df = pd.read_csv("hf://datasets/t22000t/bike-sharing-tabular/hour.csv")
print(df.shape, df["cnt"].describe())

Or via the modelling pipeline:

git clone https://github.com/timothy22000/tabular_data_modelling_pipeline
cd tabular_data_modelling_pipeline
pip install -e ".[all]"

python scripts/download_data.py --dataset bike_sharing
python train.py \
    --config configs/example_bike_sharing.py \
    --input data/bike_sharing.csv

Feature dictionary

Feature	Type	Description
instant	int	Record id (drop before training)
dteday	date	Date string (drop - use yr/mnth instead)
season	cat	1=spring, 2=summer, 3=fall, 4=winter
yr	int	0=2011, 1=2012
mnth	int	1-12
hr	int	Hour of day (0-23)
holiday	cat	0/1
weekday	cat	0=Sunday ... 6=Saturday
workingday	cat	1 if working day, 0 otherwise
weathersit	cat	1=clear, 2=mist, 3=light rain/snow, 4=heavy precipitation
temp	float	Normalised temperature in Celsius (divided by 41)
atemp	float	Normalised "feels-like" temperature (divided by 50)
hum	float	Normalised humidity (divided by 100)
windspeed	float	Normalised wind speed (divided by 67)
casual	int	Leakage - non-registered user count (excluded from features)
registered	int	Leakage - registered user count (excluded from features)
cnt	int	Target - total rentals (casual + registered)

casual and registered sum to cnt and must be excluded from the feature set. The shipped config does this.

Baseline metrics (8-architecture pipeline)

Baseline metrics will be filled in here once the model collection lands at t22000t/bike-sharing-tabular-models.

Splits

Single CSV - 17,379 hourly rows from Jan 2011 to Dec 2012. The pipeline does its own 80/20 random split (deterministic with seed=42). For a more realistic time-series split, set DatasetConfig.split_col to a column you construct (e.g. "before/after 2012-09").

Personal and sensitive information

None. Each row is a count of aggregated hourly bike rentals from the Capital Bikeshare system in Washington, DC. No individual rider data.

License and attribution

CC BY 4.0. Original publication:

Fanaee-T, Hadi, and Gama, Joao. Event labeling combining ensemble detectors and background knowledge. Progress in Artificial Intelligence (2014): pp. 1-15, Springer Berlin Heidelberg.

UCI ML Repository link: https://archive.ics.uci.edu/dataset/275/bike+sharing+dataset

Citation

@article{fanaee2014event,
  title   = {Event labeling combining ensemble detectors and background knowledge},
  author  = {Fanaee-T, Hadi and Gama, Jo{\~a}o},
  journal = {Progress in Artificial Intelligence},
  pages   = {1--15},
  year    = {2014},
  publisher = {Springer Berlin Heidelberg}
}

@software{tabular_data_modelling_pipeline,
  author = {Mun, Timothy},
  title  = {tabular-data-modelling-pipeline},
  url    = {https://github.com/timothy22000/tabular_data_modelling_pipeline},
  year   = {2026}
}

Related

• 🤖 t22000t/bike-sharing-tabular-models - pre-trained models on this dataset
• 📂 t22000t/house-prices-tabular - companion dataset (gamma family)
• 📦 tabular-data-modelling-pipeline - the underlying pipeline