time-series-transformers 0.1.0

sklearn transformers for working with time series data

time-series-transformers

scikit-learn transformers for working with time series data.

time-series-transformers is a collection of scikit-learn-compatible transformers for common time series preprocessing tasks — scaling, log and signed-power transforms, differencing, detrending, and trend extraction. Every transformer operates directly on pandas DataFrames and preserves their index and column structure, follows the familiar fit / transform API, and — where it makes sense — implements inverse_transform so you can map results back to the original scale.

Installation

pip install time-series-transformers

Usage

All transformers accept and return pandas DataFrames:

import pandas as pd

from time_series_transformers import (
    DifferenceTransformer,
    InvertiblePipeline,
    LogTransformer,
)

data = pd.DataFrame({"price": [100.0, 102.0, 101.5, 105.0, 110.0]})

# Chain transforms; InvertiblePipeline undoes them in reverse order.
pipeline = InvertiblePipeline(
    [
        ("log", LogTransformer()),
        ("difference", DifferenceTransformer()),
    ]
)

transformed = pipeline.fit_transform(data)
recovered = pipeline.inverse_transform(transformed)  # back to the original prices

The package provides:

• Scaling — PandasStandardScaler, PandasMinMaxScaler
• Transforms — LogTransformer (log / signed-power)
• Stationarity — DifferenceTransformer, DetrendTransformer, HamiltonFilterTransformer
• Composition — InvertiblePipeline, DataFrameFeatureUnion

Contributing

Contributions are welcome! This project uses pixi to manage its development environment.

1. Fork and clone the repository.
2. Install the environment:

   pixi install
   

3. Create a feature branch and make your changes, adding tests where appropriate.
4. Run the checks before opening a pull request:

   pixi run test         # run the test suite
   pixi run mypy         # type-check
   pixi run pre-commit   # lint and format
   

5. Open a pull request against main with a clear description of your change.

Please make sure the test suite passes and new code is covered by tests.

License

Released under the MIT License.