Shapelet Space Representation Library

shapelet_space is a Python library for time series analysis using shapelets. This package provides tools for shapelet discovery, dynamic time warping, and transforming time series data into shapelet space. Basic functions for transforming short time-series into a user-defined shapelet-space representation vector. Longer time-series can be transformed into a matrix where each column represents the shapelet-space vector for a rolling window of subsequences.

Paper Link

https://arxiv.org/pdf/2209.04035.pdf (accepted at IEEE BigData 2022)

Installation

pip install shapelet-space

Function Descriptions

find_shapelet_space_ts

• Description: This function finds shapelets in your time series data.
• Usage:

  find_shapelet_space_ts(time_series, flatness_param)

• Parameters:
  • time_series: Your input data.
  • flatness_param: A threshold used during shapelet discovery.
• Example:

from shapelet_space import shapelet

# Create an instance of the ShapeletSpace class
shapelet_transformer = shapelet.ShapeletSpace()
# # To initialize shaplet object with custom params
# shapelet_transformer = shapelet.ShapeletSpace(Number_of_shapelets, Shapelet_array_length, Shapelet_array)

# Define your time series data
time_series = [/* your time series data here */]

# Set a flatness parameter
flatness_param = 100000  # for example

# Generate the shapelet space representation of the time series
reps = shapelet_transformer.find_shapelet_space_ts(time_series, flatness_param)

dtw_cons_md

• Description: This function computes a similarity matrix between time series sequences using the DTW method.
• Usage:

  dtw_cons_md(sequence_1, sequence_2, window_size, metric)

• Parameters:
  • sequence_1 and sequence_2: The sequences you wish to compare.
  • window_size: Determines the constraint on how much the sequences can be stretched.
  • metric: The distance metric ('euclidean', 'manhattan', etc.).
• Example:

from shapelet_space import dtw

# Prepare your data
sequence_1 = [/* your first sequence here */]
sequence_2 = [/* your second sequence here */]
window_size = 5  # for example

# Calculate the DTW distance
distance = dtw.dtw_cons_md(sequence_1, sequence_2, window_size, 'euclidean')

License

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