tacocompression 0.9.3

Implementation of TACO: A Lightweight Tree-Based Approximate Compression Method for Time Series

taco-compression

This is the official implementation of the TACO compression algorithm for time series [1].

Installation

Please install via pip:

pip install tacocompression

Usage

There are two usage modes, one as a command line tool operating on files, the other as callables from within python.

Command Line

Please use python -m tacocompression to (de)compress your time series from the command line interface. This application takes 4 named keyword arguments and at least one additional argument specifying the input(s). These inputs are the paths to the files/directory of the (compressed) time series. You can either specify a directory from which all csv files are read...

python -m tacocompression -m=c -o=compressed time_series

... or by passing the paths to the files directly.

python -m tacocompression -m=c -o=. data/time_series_1.csv data/time_series_2.csv

Each file must contain a single column interpreted as a univariate time series. We assume row 0 is the header. The compressed data is saved in csv files as well.

The keyword arguments are

• mode (short: m) Use c to compress and d to decompress.
• output (short: o) Provide a directory to save the outputs to.
• decimals (short: d) Precision of the reconstruction, i.e., number of decimals recoverable during decompression.
• pairing_function (short: p) One of "rosenberg", "szudzik", and "cantor".

To decompress the above time series, use

python -m tacocompression -m=d -o=time_series compressed

respectively

python -m tacocompression -m=d -o=data compressed_time_series_1.csv compressed_time_series_2.csv

As a Callable

from tacocompression import compress, decompress

# 2 univariate time series, length 10 and 8:
time_series = [[0.0, 1.1, 2.2, 3.3, 4.4, 5.5, 6.6, 7.7, 8.8, 9.9], [8.8, 7.7, 6.6, 5.5, 4.4, 3.3, 2.2, 1.1]]

compressed = compress(time_series, "szudzik", 1)
reconstructed = decompress(compressed, "szudzik")

assert time_series == reconstructed

[1] Bauer, André. TACO: A Lightweight Tree-Based Approximate Compression Method for Time Series. In: Proceedings of the 14th International Conference on Data Science, Technology and Applications (DATA 2025), pages 182-190. 2025.