This is the implementation of the ATTIMO algorithm for fast mining of timeseries motifs, with probabilistic guarantees.
The inner workings and guarantees of the algorithm are described in this paper.
If you find this software useful for your research, please use the following citation:
@article{DBLP:journals/pvldb/CeccarelloG22,
author = {Matteo Ceccarello and
Johann Gamper},
title = {Fast and Scalable Mining of Time Series Motifs with Probabilistic
Guarantees},
journal = {Proc. {VLDB} Endow.},
volume = {15},
number = {13},
pages = {3841--3853},
year = {2022},
url = {https://www.vldb.org/pvldb/vol15/p3841-ceccarello.pdf},
timestamp = {Wed, 11 Jan 2023 17:06:38 +0100},
biburl = {https://dblp.org/rec/journals/pvldb/CeccarelloG22.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
First, you need to install Rust on your system. The simplest way is to visit
https://rustup.rs/ and follow the instructions there. You will need the
nightly toolchain:
curl https://sh.rustup.rs -sSf | sh -s -- --default-toolchain nightly
To install the Python wrapper, issue the following commands (preferably in a virtual environment)
pip install maturin
cd pyattimo
maturin develop --release
To install the Rust cli, you can just run
cargo install --locked --force --path .
At this point, you should have the attimo command available on your system.
There is an experimental implementation of the k-motiflets definition that you can use as follows
import pyattimo
# load a dataset, any list of numpy array of floats works fine
# The following call loads the first 100000 points of the ECG
# dataset (which will be downloaded from the internet)
ts = pyattimo.load_dataset('ecg', 100000)
# Now we can find k-motiflets:
# - w is the window length
# - support is the number of subsequences in the motiflet (k in the motiflet paper)
# - repetitions is the number of LSH repetitions
m = pyattimo.motiflet(ts, w=1000, support=5, repetitions=512)
# The motiflet object allows to:
# - get the indices of the subsequences
m.indices
# - get the extent of the motiflet
m.extent
# - plot the motiflet, showing it in a window or returning the
# plot object (default) for embedding in a notebook
m.plot(show=False)
# - get the values of the i-th subsequence in the motiflet
m.values(2) # for the second subsequence
# - get the z-normalized values of the i-th subsequence in the motiflet
m.zvalues(2) # for the second subsequenceExecuting the command with no arguments shows a short help message.
attimo works with univariate time series in a very common format: text files
with one value per line:
1.1292
1.1096
1.0986
1.0925
1.0889
1.0815
1.0767
1.073
1.0681
1.0608
You can find some sample data files here.