Benj-samurai/strava_dataset

Strava Master Dataset ― Benj-samurai

Personal multi-sport training log exported from Strava (JP CSV) and processed into a clean, analysis-ready Parquet table + weekly summaries.
Covers 〔開始日 – 終了日〕, total 〔活動件数〕 activities across 〔種目数〕 sports.

---

Files

File	Rows	Description
my_strava_dataset/strava_master_enhanced.parquet	〔n〕	Master table — 1 row = 1 activity
my_strava_dataset/weekly_sport.parquet	〔m〕	Weekly totals by year–week × sport
my_strava_dataset/weekly_category.parquet	〔k〕	Weekly totals by intensity zone

(Parquet → compact, schema-aware, loadable via datasets.load_dataset)

---

Column definitions (master)

Column	dtype	Description
activity_id	int64	Strava Activity ID
name	string	Activity title as saved on Strava
sport	category	Sport type (Run, Ride, Swim, Walk, …)
date	datetime64[ns]	Local activity start time
distance_km	float32	Distance in kilometres (raw m → km)
elapsed_hr	float32	Elapsed time incl. pauses hours (raw sec → h)
moving_hr	float32	Moving time (in-motion) hours
elevation_gain_m	float32	Total positive elevation gain (m)
elevation_loss_m	float32	Total negative elevation (m)
average_speed_kph	float32	Moving speed (km h⁻¹)
max_speed_kph	float32	Max speed (km h⁻¹)
average_hr	float32	Avg heart-rate (bpm) – NaN if no sensor
max_hr	int16	Max heart-rate (bpm)
average_cadence	float32	Avg cadence (rpm)
max_cadence	int16	Max cadence (rpm)
average_power	float32	Avg power (W); bike only
max_power	int16	Peak power (W)
calories_kcal	float32	Calories reported by Strava
training_category	category	HR zone label Z1-2 / Z3 / Z4 / Z5 / NoHR
intensity_level	float32	Avg HR ÷ LTHR (165 bpm)
trimp	float32	Banister TRIMP (moving_hr × intensity_level × 50)
commute	boolean	Marked as commute on Strava
filename	string	Original FIT/GPX filename (meta only)
gear	string	Bike / shoes used (if set)
weather	string	Weather summary (if available)
temperature_c	float32	Avg temp (°C)
flagged	boolean	Strava flagged activity
year	int16	Calendar year (date). fast grouping
month	int16	Calendar month (1–12)
week	int16	ISO week number (1–53)
year_month	string	"YYYY-MM" label for plotting
week_start	datetime64[ns]	Monday of ISO week (analysis helper)
sport_weekly_id	string	Composite key year_week-sport
distance_ratio	float32	Share of weekly distance (per sport)
pace_min_per_km	float32	Pace (min km⁻¹); NaN for non-run
grade_adjusted_pace	float32	GAP (min km⁻¹); run only
dirt_distance_km	float32	Unpaved distance (km)
total_cycles	int32	Swim strokes / pedal revs where available
route_hash	string	MD5 of polyline (GPS privacy)
gpx_path	string | None	Optional GeoJSON path file

All numeric distance/time columns are converted to km / hours and stored in low-memory float32/int16 where possible.
Empty sensor data are kept as NaN so they don’t skew means.

---

Processing pipeline

1. Export: Strava JP CSV (activities.csv)
2. Header translation JP→EN (translate_headers.py)
3. Unit conversion (m→km, s→h) & dtype down-cast (clean_master.ipynb)
4. Intensity & TRIMP: LTHR = 165 bpm, Banister formula
5. Weekly aggregations (weekly_summary.ipynb)
6. Saved as Parquet, tracked via Git LFS (compact & diff-friendly)

Code & notebooks live in the companion GitHub repo:
https://github.com/Benj-samurai/Strava-Dataset-PRJ

---

Usage ✨

from datasets import load_dataset

ds = load_dataset(
    "Benj-samurai/strava_dataset",
    data_files="my_strava_dataset/strava_master_enhanced.parquet",
    streaming=False,    # True = stream without download
)
df = ds["train"].to_pandas()

# quick EDA
weekly_km = (
    df.assign(year_week=df["date"].dt.to_period("W"))
      .groupby(["year_week", "sport"])["distance_km"].sum()
)
print(weekly_km.tail())

Privacy & Personal-use License

• Raw FIT/GPX files are not included.
• Activity start coordinates are jittered ≥ 200 m to obscure the true home location.
• Released under CC BY-NC 4.0 – non-commercial use, attribution required.
  If you wish to use the data commercially, please contact the author first.

---

Citation

@misc{asai2025strava,
  author       = {Asai, Benj-samurai},
  title        = {Strava Master Dataset},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/Benj-samurai/strava_dataset}},
  note         = {Version {{\today}}}
}

Changelog

Date	Version	Notes
2025-05-06	v1.0	Initial public release

---

使い方

1. HF Hub ページの Dataset card タブ → Edit を開く
2. 上の Markdown を貼り付ける
3. 〔 〕部分を実際の値に置換して Commit
   行数 は手元で len(df)、週レコードは len(weekly_sport) などで確認できます。

これで “列定義・処理手順・使用例・ライセンス” を網羅したリッチな Dataset Card になります。
追記やレイアウト調整はお好みでどうぞ！