PYMTASA (PYthon Multivariate Time series Alignment and Similarity Assessment)

Overview

PYMTASA is a Python package designed for Multivariate Time series Alignment and Similarity Assessment. This package provides a framework for computing similarity index matrices for multivariate time series data. It's particularly useful for scenarios where temporal shifts need to be considered in the similarity assessment.

Installation

You can install the dependencies using pip:

pip install -r requirements.txt

Input data

• measurement_vars (tuple): Names of the measurement variables to use, e.g. ("prec","tmean").
• weights (tuple): Weights for each variable in the range 0-1. The sum of the weights must equal 1, e.g., (0.5, 0.5).
• number_divisions (tuple): Number of divisions for each variable, e.g., (12, 12).
• ref_data (string): File path of the query sequence data.
• target_dataset (tuple): File paths for the time series dataset.
• headers_indices (list): List of headers indices corresponding to the measurement values within the time series files.
• analysis_period (list): Period of interest in the analysis.
• rotation_variables (list): List of rotation variables used for comparing instances with temporal shifts.
• rotation_mode (bool): Specify if the rotation is absolute or relative.
• threshold (float): Similarity threshold value between 0 and 1.
• threshold_mode (bool): Specify if the threshold is absolute or relative.
• outfile (string): Directory where the resultant similarity matrix will be saved.
• file_name (string): Name of the output file.
• write_file (boolean): Whether to write the output file to disk.

USAGE

Here's an example of how to use PYMTASA:

from pymtasa.parameters_set import ParametersSet
from pymtasa.similarity import Similarity
from pymtasa.static_variables import RESULTS_DIRECTORY, TIME_SERIES_DATASET, QUERY_SEQUENCE, HEADERS_INDICES

if __name__ == '__main__':
    mtasa_parameters = ParametersSet(
        measurement_vars=("prec", "tmean"),
        weights=(0.5, 0.5),
        number_divisions=(12, 12),
        ref_data=QUERY_SEQUENCE,
        target_dataset=TIME_SERIES_DATASET,
        headers_indices=HEADERS_INDICES,
        analysis_period=[11, 12],
        rotation_variables=["prec", "tmean"],
        threshold=0,
        rotation_mode=True,
        threshold_mode=True,
        outfile=RESULTS_DIRECTORY,
        file_name="pyresults",
        write_file=True
    )

    similarity = Similarity(mtasa_parameters)

    similarity_index_matrix = similarity.compute_similarity_matrix()

Results

The package will compute a similarity index matrix (see an example below) based on the provided parameters. The results can be saved to the specified directory.

ID	Rotation coefficient	Similarity index
1	5	0.96
2	7	0.75
...	...	...

License

This project is licensed under the MIT License - see the LICENSE file for details.