Skip to content
Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Data sets

This folder contains the data sets that we used to obtain the results reported in the paper. All data have been obtained from the MIMIC Critical Care Database; please refer to the paper for more details.


  • training: contains the files we used for training our method, i.e., the data that we used for extracting statistically significant shapelets in the first place
  • test: contains the files we used for testing the results of our method in the context of classifying time series

In the following, we will describe some operations that you can apply to the data for your application.


The blood pressure data has been downsampled because of the large number of very similar time series with respect to the cases and controls. The training folder contains the sample that we used to obtain the results in the paper. Please follow these steps in your preferred terminal emulator to create your own subsampled version of the data (assuming that you have access to a POSIX-like operating system such as Linux or MacOS X):

# First, we need to extract cases and controls from the input data,
# so we have to take a look at the first entry of each row.
$ grep    "^1" Blood_pressure.csv > Blood_pressure_cases.csv
$ grep - v "^1" Blood_pressure.csv > Blood_pressure_controls.csv
# We then bring all cases into random order, take 50 of them, which
# you can of course change, and output them to a new file. Ditto for
# the controls, which are appended to the new file.
$ sort -R Blood_pressure_cases.csv    | head -n 50 >  Blood_pressure_random_sample.csv
$ sort -R Blood_pressure_controls.csv | head -n 50 >> Blood_pressure_random_sample.csv
# Last, as a good data science tactic, we shuffle the resulting data
# set in order to ensure that the order in which we specified series
# does not matter. This is not relevant for our method, but may turn
# out to be relevant for comparing with other methods.
$ gshuf -o Blood_pressure_random_sample.csv < Blood_pressure_random_sample.csv

Reproducing the results

Assuming that you have successfully built our method in a build subdirectory of the repository, the following settings have been used to create the results:

$ ./s3m -m 4 -M 6 -t ../data/training/Blood_pressure_random_sample > /tmp/Blood_pressure.json
$ ./s3m -m 4 -M 6 -t ../data/training/Heart_rate.csv               > /tmp/Heart_rate.json
$ ./s3m -m 4 -M 6 -t ../data/training/Respiratory_rate.csv         > /tmp/Respiratory_rate.json

You can experiment with other settings of the s3m executable in order to obtain longer or shorter shapelets, for example. Please use the -h or --help flag of the executable for a brief description:

$ ./s3m -h
You can’t perform that action at this time.