wangdx25/monash_uea_ucr_tser

Dataset Card for Time Series Extrinsic Regression

Dataset Summary

A collection of datasets from Monash, UEA, and UCR supporting research into Time Series Extrinsic Regression (TSER), a regression task of which the aim is to learn the relationship between a time series and a continuous scalar variable. This task is closely related to time series classification, where a single categorical variable is learned. Please read the paper for more.

If you use the results or code, please cite the paper "Chang Wei Tan, Christoph Bergmeir, Francois Petitjean, Geoffrey I. Webb, Time Series Extrinsic Regression: Predicting numeric values from time series data". (Full BibTex citation can be found at the end of this card).

(It has been generated using this raw template.)

Supported Tasks and Leaderboards

[More Information Needed]

Languages

Dataset Structure

Data Instances

A sample from the training set of Appliances Energy (a multivariate time series dataset) is provided. The following is a single record from that dataset:

{'start': Timestamp('2016-02-28 17:00:00'),
 'feat_static_cat': 0,
 'to_predict': 19.38,
 'timeseries': array([[21.29      , 21.29      , 21.29      , ..., 21.79      ,
         21.79      , 21.79      ],
        [31.66666667, 31.92666667, 32.06      , ..., 33.66      ,
         33.7       , 33.56666667],
        [19.89      , 19.82333333, 19.79      , ..., 19.79      ,
         19.79      , 19.79      ],
        ...,
        [ 7.        ,  6.83333333,  6.66666667, ...,  5.        ,
          5.        ,  5.        ],
        [40.        , 40.        , 40.        , ..., 40.        ,
         40.        , 40.        ],
        [-4.2       , -4.16666667, -4.13333333, ..., -4.3       ,
         -4.16666667, -4.03333333]]),
 'item_id': 'item_000'}

Data Fields

This format was loosely adapted from the Gluon format and the HF convention also seen in the recent series of Time Series Transformer notebooks

• start: a datetime of the first entry of each time series in the data record
• feat_static_cat: the original identifier given to this record
• timeseries: the timeseries itself
• to_predict: continuous variable to predict
• item_id: an identifier given to each record (for e.g. group-by style aggregations)

The timeseries field will be a single array in the univariate forecasting scenario, and a 2-D array in the multivariate scenario.

The to_predict will be a single number in most cases, or an array in a few instances (noted in the table above TODO).

Data Splits

Train and test are temporally split (i.e. "train" is the past and "test" is the future) 70/30 whenever possible, though some datasets have more particular splits.

For details, see the paper and the particular dataset you are interested in. In our porting to HF Hub, we made as few changes as possible.

Dataset Creation

While I (Stephen) did not create the original dataset, I took the initiative to put the data on Hugging Face Hub. Any grievances with the dataset should first and foremost be directed to me.

Curation Rationale

To facilitate the evaluation of global forecasting models that are predicting a single-point estimate in the future. All datasets in the repository are intended for research purposes and to evaluate the performance of new TSER algorithms. This

Source Data

Initial Data Collection and Normalization

The origins of each dataset are articulated in the paper.

Minimal preprocess was applied to the dataset, as they are still in their sktime-compatible .ts format. (As far as Stephen is aware.)

Who are the source language producers?

The data comes from the datasets listed in the paper and in the table on the website

Annotations

Annotation process

Please see the paper for the annotation aggregation propcess

Who are the annotators?

The annotation comes from the datasets listed in the paper and in the table on the website

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

• Chang Wei Tan
• Anthony Bagnall
• Christoph Bergmeir
• Daniel Schmidt
• Eamonn Keogh
• François Petitjean
• Geoff Webb

Licensing Information

GNU General Public License (GPL) 3

Citation Information

@article{
  Tan2020TSER,
  title={Time Series Extrinsic Regression}, 
  author={Tan, Chang Wei and Bergmeir, Christoph and Petitjean, Francois and Webb, Geoffrey I},
  journal={Data Mining and Knowledge Discovery},
  pages={1--29},
  year={2021},
  publisher={Springer},
  doi={https://doi.org/10.1007/s10618-021-00745-9}
}

Contributions

[More Information Needed]

TsFile Conversion

• Original dataset: foxy-steve/monash_uea_ucr_tser

• Modalities: Time-series

• Converted data files are listed in the YAML metadata above.

• Source README text and dataset-specific metadata are retained; the source Usage section is replaced with the executable TsFile Python SDK example below.

• start anchors each series and is encoded as millisecond Time with one millisecond per positional sample because the source card does not provide a sampling frequency.

• The 2-D timeseries array is expanded to timeseries_00, timeseries_01, ... FLOAT fields; item_id, feat_static_cat, and split are TAG columns.

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("monash_uea_ucr_tser_test.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())