This Python package lets you compute the Iterated Sums Signature of a time series in numpy. All
methods are implemented using numba, which leads to a large speed-up of computations through
compiling functions directly in Python.
Together with weighted iterated sums and iterated sums over different semirings, this repository provides a toolbox for time series feature extraction. This approach has been already explored in our publication of FRUITS.
The package is installable via pip:
$ pip install witss
The project's dependecies are just numpy and numba.
For development or further testing, it is recommended to use poetry
or install the development dependencies listed in pyproject.toml otherwise.
Importing the package for the first time could take a minute, as some functions are compiled by
numba and then cached for later use.
Examples can be found in /examples. For running the examples, you also have to install the development dependencies.