Datasets:

THULab
/

dysts

pretty_name: Dysts Chaotic Dynamical-System Trajectories (TsFile)
modality: timeseries
license: cc-by-4.0
authors: William Gilpin
task_categories:
  - time-series-forecasting
size_categories:
  - 1M<n<10M
tags:
  - tsfile
  - timeseries
  - modality:timeseries
  - format:tsfile
  - dynamical-systems
  - chaotic-systems
  - simulation
configs:
  - config_name: coarse
    data_files:
      - split: coarse
        path: dysts_coarse_*.tsfile
  - config_name: medium
    data_files:
      - split: medium
        path: dysts_medium_*.tsfile
  - config_name: fine
    data_files:
      - split: fine
        path: dysts_fine_*.tsfile

Dysts Chaotic Dynamical-System Trajectories (TsFile)

This dataset is the Apache TsFile conversion of williamgilpin/dysts, a collection of simulated trajectories from named chaotic and nonlinear dynamical systems. See the source papers arXiv:2110.05266 and arXiv:2303.08011.

Modalities: Time-series.

Overview

Source revision: 5d305abb3c9010e36ca6a3e6c967b72932f6cca9
Source files: coarse/*.csv, medium/*.csv, and fine/*.csv
Trajectories/files: 352 total (117 coarse, 118 medium, 117 fine)
Points per trajectory: 10,000
Converted rows: 3,520,000 total (1,170,000 coarse, 1,180,000 medium, 1,170,000 fine)
License: cc-by-4.0

Each CSV contains a time column and a variable number of numeric channels. The source channel names are commonly x0, x1, ...; the leading whitespace is normalized during conversion. Different systems may have different channel counts, so each output table uses a union schema.

TsFile files and schema

The converter emits one logical table per source resolution. Each logical table is split into two TsFile shards by the import tool:

File	Table	Rows	Size (bytes)
`dysts_coarse_1.tsfile`	`dysts_coarse`	1,048,576	29,477,332
`dysts_coarse_2.tsfile`	`dysts_coarse`	121,424	3,581,515
`dysts_medium_1.tsfile`	`dysts_medium`	1,048,576	30,225,699
`dysts_medium_2.tsfile`	`dysts_medium`	131,424	3,893,846
`dysts_fine_1.tsfile`	`dysts_fine`	1,048,576	30,972,948
`dysts_fine_2.tsfile`	`dysts_fine`	121,424	3,527,556

Every table has the following common columns:

Column	Role	Type	Meaning
`Time`	TIME	INT64 (ms)	Rounded source `time * 1000`; source values run from 0 to 9,999
`granularity`	TAG	STRING	`coarse`, `medium`, or `fine`
`system`	TAG	STRING	System name parsed from the source filename
`x0` through `x9`	FIELD	DOUBLE	Numeric trajectory channels; unavailable channels are null

The actual number of populated channel fields depends on the system. The union schema keeps the maximum observed channel names (x0 through x9) in each resolution table.

Conversion notes

Source time is interpreted as seconds and converted to integer-millisecond Time by rounding. It is not retained as a second field.
The filename suffix (_coarse, _medium, _fine) supplies the granularity TAG; the remaining filename stem supplies system.
Channel names are trimmed and sanitized to TsFile-safe identifiers. Numeric values are copied without scaling. Missing channels in a system are null; no values are imputed and no trajectory rows are dropped.
Files are processed deterministically and sorted by granularity, system, and Time. Each source system has unique (system, Time) keys within its resolution.

Read example

from tsfile import TsFileReader

path = "dysts_coarse_1.tsfile"
with TsFileReader(path) as reader:
    with reader.query_table(
        "dysts_coarse",
        ["x0", "x1", "x2"],
        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/williamgilpin/dysts
Author / publisher: William Gilpin
Papers: https://arxiv.org/abs/2110.05266 and https://arxiv.org/abs/2303.08011
License: CC-BY-4.0

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("dysts_coarse_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())