Home

Awesome

<div style="text-align:center"><img src="docs/mass-logo.png" /></div>

Introduction

Tweet

<img src="https://img.shields.io/pypi/v/mass_ts.svg"> <img src="https://img.shields.io/travis/tylerwmarrs/mass-ts.svg">

MASS allows you to search a time series for a subquery resulting in an array of distances. These array of distances enable you to identify similar or dissimilar subsequences compared to your query. At its core, MASS computes Euclidean distances under z-normalization in an efficient manner and is domain agnostic in nature. It is the fundamental algorithm that the matrix profile algorithm is built on top of.

mass-ts is a python 2 and 3 compatible library.

Free software: Apache Software License 2.0

Features

Original Author's Algorithms

Library Specific Algorithms

Installation

pip install mass-ts

GPU Support

Please follow the installation guide for CuPy. It covers what drivers and environmental dependencies are required. Once you are finished there, you can install GPU support for the algorithms.

pip install mass-ts[gpu]

Example Usage

A dedicated repository for practical examples can be found at the mass-ts-examples repository.


import numpy as np
import mass_ts as mts

ts = np.loadtxt('ts.txt')
query = np.loadtxt('query.txt')

# mass
distances = mts.mass(ts, query)

# mass2
distances = mts.mass2(ts, query)

# mass3
distances = mts.mass3(ts, query, 256)

# mass2_gpu
distances = mts.mass2_gpu(ts, query)

# mass2_batch
# start a multi-threaded batch job with all cpu cores and give me the top 5 matches.
# note that batch_size partitions your time series into a subsequence similarity search.
# even for large time series in single threaded mode, this is much more memory efficient than
# MASS2 on its own.
batch_size = 10000
top_matches = 5
n_jobs = -1
indices, distances = mts.mass2_batch(ts, query, batch_size, 
    top_matches=top_matches, n_jobs=n_jobs)

# find minimum distance
min_idx = np.argmin(distances)

# find top 4 motif starting indices
k = 4
exclusion_zone = 25
top_motifs = mts.top_k_motifs(distances, k, exclusion_zone)

# find top 4 discord starting indices
k = 4
exclusion_zone = 25
top_discords = mts.top_k_discords(distances, k, exclusion_zone)

Citations

Abdullah Mueen, Yan Zhu, Michael Yeh, Kaveh Kamgar, Krishnamurthy Viswanathan, Chetan Kumar Gupta and Eamonn Keogh (2015), The Fastest Similarity Search Algorithm for Time Series Subsequences under Euclidean Distance, URL: http://www.cs.unm.edu/~mueen/FastestSimilaritySearch.html