THULab/timeseries_finance_etf_missing

Time Series Finance ETF (TsFile)

This dataset is the Apache TsFile conversion of mik3ml/timeseries-finance-ETF. The source is a JSON Lines collection of financial forecasting windows: each record contains an id, an ISO-8601 timestamp array, a matching target array, category, and Ticker.

Modalities: Time-series.

Overview

• Source revision: 5ccaaef9d9477def3321b9c83e60ca1709f6a499
• Source file: dataset-etf.jsonl
• Source records/windows: 41,850
• Converted observations: 2,406,262 scalar rows
• Converted files: 3 TsFile shards
• Split: train

The source row is a forecasting window rather than one scalar observation. Timestamp spacing follows the source ETF calendar (including its trading-day gaps); no regular-frequency values are synthesized.

TsFile layout

The converter writes the logical table timeseries_finance_etf_missing in three shards:

File	Rows	Size (bytes)
timeseries_finance_etf_missing_1.tsfile	1,048,576	11,378,556
timeseries_finance_etf_missing_2.tsfile	1,048,576	11,385,511
timeseries_finance_etf_missing_3.tsfile	309,110	3,339,111

Rows are sorted by series_id, ticker, category, Time, and point_index.

Column	Role	Type	Source / meaning
Time	TIME	INT64 (ms)	UTC epoch milliseconds parsed from each source timestamp string
series_id	TAG	STRING	Source id, identifying the forecasting window
ticker	TAG	STRING	Source Ticker
category	TAG	STRING	Source category
value	FIELD	FLOAT	One scalar from the source target array
point_index	FIELD	INT64	Zero-based position within the source window

Conversion notes

• timestamp and target arrays are expanded in lockstep into one row per point. Rows with mismatched array lengths are rejected instead of being silently truncated.
• timestamp is parsed as UTC and encoded as Time in integer milliseconds; the original array is not retained as a nested column.
• target is stored as single-precision value; point_index is added so the original order within a window remains explicit.
• id, Ticker, and category are preserved as the series_id, ticker, and category TAGs. This keeps windows with otherwise equal dates in separate TsFile devices.
• No source measurement values are imputed or otherwise changed. The nested JSON representation is the only structural normalization.

Read example

from pathlib import Path
from tsfile import TsFileReader

# Use any of the three generated shards.
path = next(Path(".").glob("**/timeseries_finance_etf_missing*.tsfile"))
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print(list(schemas))
    with reader.query_table(
        "timeseries_finance_etf_missing",
        ["value", "point_index"],
        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/mik3ml/timeseries-finance-ETF
• Author / publisher: mik3ml
• License: Apache-2.0

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("timeseries_finance_etf_missing_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())