CORAD: Correlation-Aware Compression of Massive Time Series using Sparse Dictionary Coding

• Overview: CORAD is a new real-time technique to effectively compress time series streams. It relies on a dictionary-based technique that exploits the correlation across time series. In addition, CORAD allows to adjust the degree of accuracy that is acceptable depending on the use-case. Technical details can be found in our Big Data 2019 paper: CORAD: Correlation-Aware Compression of Massive Time Series using Sparse Dictionary Coding .

• Compression steps:

  1. Create the tricklets
  2. Compute the subsequence correlations
  3. Learn the dictionary
  4. Sparse code the data

• Datasets: We use datasets from the UCR Time Series Classification Archive (UCR), the UCI Machine Learning Repository (UCI), and the Swiss Federal Office for the Environment(FOEN). All the datasets used in the paper can be found here.

Prerequisites | Build | Execution | Arguments | Contributors | Citation

---

Prerequisites

• Ubuntu 18 and 20 (including the same distribution under WSL) or Mac OS.
• Clone this repo

---

Build

• To install all the dependencies, run the following installation script:

    $ sh install.sh

---

Execution

    $ python3 corad.py [args]

Arguments

args	Interpretation
-d	Name of the dataset (comma-separated-values, tabular-seperated-values, etc.)
-t	Length of the tricklets
-e	Max loss between the original data and the compressed one
-a	Number of atoms used for the representation of each tricklet

Execution Examples

1. Compress the PigAirwayPressure dataset with the default parameters (trick=40, err=0.4, atoms=4)

python3 corad.py -d 'datasets/PigAirwayPressure_TEST.tsv'

2 . Compress the PigAirwayPressure dataset with a customized error threshold

python3 corad.py -d 'datasets/PigAirwayPressure_TEST.tsv' -e 0.1

3 . Compress the PigAirwayPressure dataset with customized error threshold, and number of atoms

python3 corad.py -d 'datasets/PigAirwayPressure_TEST.tsv' -e 0.1 -a 6

4 . Compress the PigAirwayPressure dataset with customized tricklets length, error threshold, and number of atoms

python3 corad.py -d 'datasets/PigAirwayPressure_TEST.tsv' -t 20 -e 0.1 -a 6

Results

All the results including the compressed data, runtime, accuracy error, and the compression ratios will be added to results/{dataset_name}.txt file. The results of the baseline TRISTAN technique will be also added.

The compressed data are exported using Python's pickle library into the results/compressed_data/{dataset}/ folder and could be opened using the following command:

Decompression

In addition to data decompression in the main scripts, decompression.py allows to perform a decompression as follows:

python3 corad.py -d 'datasets/PigAirwayPressure_TEST.tsv'

python3 decompress.py -d 'results/compressed_data/PigAirwayPressure_TEST.tsv/originalData.out'

---

Contributors

• Abdelouahab Khelifati (abdel@exascale.info)
• Dr. Mourad Khayati (mourad@khayati.unifr.ch)

---

Citation

@inproceedings{DBLP:conf/bigdataconf/KhelifatiKC19,
  author    = {Abdelouahab Khelifati and
               Mourad Khayati and
               Philippe Cudr{\'{e}}{-}Mauroux},
  title     = {{CORAD:} Correlation-Aware Compression of Massive Time Series using
               Sparse Dictionary Coding},
  booktitle = {2019 {IEEE} International Conference on Big Data (Big Data), Los Angeles,
               CA, USA, December 9-12, 2019},
  pages     = {2289--2298},
  publisher = {{IEEE}},
  year      = {2019},
  url       = {https://doi.org/10.1109/BigData47090.2019.9005580},
  doi       = {10.1109/BigData47090.2019.9005580},
  timestamp = {Fri, 09 Apr 2021 17:11:11 +0200},
  biburl    = {https://dblp.org/rec/conf/bigdataconf/KhelifatiKC19.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}