eral 0.0.3

Implementation of ERAL algorithm (Stržinar et al., 2024)

ERAL Algorithm

Introduction

The ERAL (Error In Alignment) algorithm is a state-of-the-art method designed for time series alignment and averaging. The method obtains the average time series (the prototype) from a set of time series (a class).

Installation

To use ERAL, please clone the git repository:

git clone https://repo.ijs.si/zstrzinar/eral.git

The repository contains a requirements file, ensure you have all the requirements

pip install -r requirements.txt

pip package

A pip package is being prepared and will be made available soon.

Usage

The main entry point for the end user is obtain_prototype method in eral.eral:

from src.eral.eral import obtain_prototype

class_prototype = obtain_prototype(...)

The function takes several arguments, most important are:

• class_segments: a list of time series to be aligned and averaged
• prototyping_function: determines how the average is computed, typically the function get_new_prototype_variable_clipping from eral.alignment_prototyping_functions can be used.
• exclustion_zone: percentage of forbidden alignments (see [1])

The basic function call is:

import numpy as np
from src.eral.eral import obtain_prototype
from src.eral.alignment_prototyping_functions import get_new_prototype_variable_clipping

X: list[np.ndarray] = [...]  # Class data

class_prototype = obtain_prototype(X,
                                   prototyping_function=get_new_prototype_variable_clipping,
                                   exclusion_zone=0.2)

For full examples, please refer to the examples/ directory.

Examples

The examples/ directory contains Jupyter notebooks that illustrate different uses and capabilities of the ERAL algorithm. To run an example, navigate to the examples/ directory and execute the desired notebook.

• Notebook titled 01 ERAL demo demonstrates the ERAL prototyping method using the Trace dataset from UCR Archive.
• Notebook titled 02 ERAL demo on industrial data downloads an industrial dataset from Mendeley Data, and calculates the prototypes for all classes.
• Notebook titled 03 Comparison compares ERAL to DBA, SSG and others, using implementations in tslearn

References

[1] Paper introducing ERAL [2] Stržinar, Žiga; Pregelj, Boštjan; Petrovčič, Janko; Škrjanc, Igor; Dolanc, Gregor (2024), “Pneumatic Pressure and Electrical Current Time Series in Manufacturing”, Mendeley Data, V2, doi: 10.17632/ypzswhhzh9.2, url: https://data.mendeley.com/datasets/ypzswhhzh9/2