THULab/weather_data_bavaria

Bavaria Weather Data (TsFile)

This dataset is the Apache TsFile conversion of Kamilatr/weather_data_bavaria. It is a large station-and-date weather table for Bavaria and surrounding stations, delivered by the source as one CSV file.

Modalities: Time-series.

Overview

• Source dataset: Kamilatr/weather_data_bavaria
• Source revision: ff3fe5baa843f02c9aaf4328c567779cd44d0e66
• Source file: bavaria_weather_data_2024-08-14.csv
• Rows/observations: 8,867,345
• Converted files: 89 TsFile shards (261,160,977 bytes total)
• Distinct station_id values: 232
• Source month partitions: 1 through 12
• Timestamp range: 1940-01-01 through 2024-08-14
• Split: train

The source card does not declare a license. The table is retained as supplied, including its irregular station coverage and missing meteorological values.

TsFile schema

The logical table is weather_data_bavaria. station_id and the source month partition are the device TAGs. The import tool produced 89 shards, all with the same schema.

Column	Role	Type	Meaning
Time	TIME	INT64 (ms)	UTC-midnight epoch milliseconds from timestamp
station_id	TAG	STRING	Source station identifier (numeric and alphanumeric IDs are preserved)
month	TAG	STRING	Source month partition value, preserved verbatim
precipitation, threshold, exceedance	FIELD	DOUBLE	Source precipitation-related measurements
avg_temp, min_temp, max_temp	FIELD	DOUBLE	Source temperature measurements
wind_dir, wind_speed, wind_peak_gust	FIELD	DOUBLE	Source wind measurements
sea_level_pressure	FIELD	DOUBLE	Source pressure measurement
snow_depth, sunshine_duration	FIELD	DOUBLE	Source snow/sunshine measurements

Conversion notes

• Conversion is streamed in 100,000-row CSV chunks, then written to a staged Parquet file and imported to TsFile. No rows are intentionally filtered.
• timestamp is parsed as a calendar date and encoded as UTC-midnight Time in integer milliseconds. The redundant source string is not duplicated as a FIELD.
• station_id is read as a string so identifiers such as NHP5I are not coerced to numbers. month remains the source partition string rather than being recomputed.
• All twelve measurement fields are retained. Nulls are preserved as TsFile nulls; no interpolation, filling, or unit conversion is applied.

Read example

from pathlib import Path
from tsfile import TsFileReader

path = next(Path(".").glob("**/weather_data_bavaria*.tsfile"))
with TsFileReader(str(path)) as reader:
    print(reader.get_all_table_schemas().keys())
    with reader.query_table(
        "weather_data_bavaria",
        ["precipitation", "avg_temp", "sea_level_pressure"],
        batch_size=4096,
    ) as result:
        batch = result.read_arrow_batch()
        if batch is not None:
            print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/Kamilatr/weather_data_bavaria
• Author / publisher: Kamilatr
• License: not declared by the original dataset; please defer to the original.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("weather_data_bavaria_1.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())