smilegeng/weather_data_NC

Weather Data — North Carolina (hourly, TsFile)

This dataset is a lossless conversion to the Apache TsFile format of the HuggingFace dataset Katherinetian/weather_data_NC: hourly OpenWeather observations for 9 cities in North Carolina, USA.

Original dataset

• Source dataset: Katherinetian/weather_data_NC
• Data origin: OpenWeather API
• Content: hourly weather (temperature, pressure, humidity, wind, clouds, precipitation, weather condition codes) for 9 NC cities.

Scale

• 78,615 rows = 9 cities × 8,735 hourly readings (after de-duplication; see below)
• 22 source columns → 21 stored (the date column is dropped)
• Time range: 2023-03-19 04:00 → 2024-03-17 03:00 (UTC), hourly cadence
• The 9 cities share one time axis.

TsFile storage mapping (table model)

Role	Column(s)	Type	Notes
TAG	city	STRING	9 cities; city ↔ (latitude, longitude) is 1:1, one city = one device
Time	source dt	INT64 (ms)	source is Unix epoch seconds, converted to milliseconds
FIELD	main_temp, main_feels_like, main_temp_min, main_temp_max, wind_speed, wind_gust, latitude, longitude, rain_1h, snow_1h, rain_3h	DOUBLE	source float64
FIELD	main_pressure, main_humidity, wind_deg, clouds_all, weather_id	INT64	source int64
FIELD	weather_main, weather_description, weather_icon	STRING	weather condition labels

(Source column names contain dots, e.g. main.temp; dots are replaced with underscores to form TsFile-safe identifiers, e.g. main_temp.)

Conversion notes

• TAG = city (9 devices). Required so each device's time axis is strictly increasing.
• Time: source dt (Unix epoch seconds) → INT64 epoch milliseconds. Rows sorted ascending by (city, Time).
• Dropped column (with consent): date — a redundant date-string of dt (~18% misaligned by timezone). Its information is preserved losslessly in Time. This is the only column dropped.
• De-duplication (lossless): the source contains 9 duplicate (city, dt) keys (all at dt=1710129600, one per city). Within each duplicated pair the only differing column is date (which is dropped anyway), so the first row of each pair is kept and the duplicate is removed — 78,624 → 78,615 rows, no field values lost. This is required because TsFile forbids duplicate/out-of-order timestamps within a device.
• Nulls kept as-is: several precipitation/gust columns are mostly empty (rain_3h and snow_1h are ~100% null, rain_1h ~95%, wind_gust ~63%). They are preserved — TsFile simply does not write null cells. No rows dropped (beyond the 9 exact-duplicate removals above).
• Note on a city value: one device is labelled Winnabow,NC,USChapel Hill,NC,US (two city names concatenated in the source). This is a source-data glitch and is kept verbatim, not rewritten.

Layout

data/
└── weather_data_nc.tsfile

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/weather_data_nc.tsfile")
schemas = reader.get_all_table_schemas()
tname = next(iter(schemas))

cols = ["city", "main_temp", "main_humidity", "wind_speed", "weather_main"]
with reader.query_table(tname, cols, batch_size=65536) as rs:
    while (batch := rs.read_arrow_batch()) is not None:
        df = batch.to_pandas()
        # ... process ...
reader.close()

Citation

@dataset{weather_data_nc,
  title     = {Weather Data — North Carolina},
  author    = {Katherinetian},
  url        = {https://huggingface.co/datasets/Katherinetian/weather_data_NC},
  publisher = {Hugging Face}
}

Weather data sourced from the OpenWeather API; please follow OpenWeather's terms for downstream use. The source HuggingFace dataset does not declare an explicit license.