Amerced DTW demonstration application

Amerced DTW (ADTW) is a variant of DTW, "amercing" ("penalizing") warping steps by a fixed penalty. This is a NN1 classification demonstration application for our paper (submitted)

Amercing: An Intuitive, Elegant and Effective Constraint for Dynamic Time Warping

This program can load ts files from the UCR128 archive (see timeseriesclassification.com).

UCR128 Classification Results

The results folder contains 10 csvs -- one for each exponent p=1..10, using 2000 samples for the maximum penalty omega'. A csv contains a line per dataset from the UCR128 archive, excluding datasets with varying length or missing data (hence 113 remaining).

Note: results presented in the paper are based in 112 datasets, not 113. This is because the Fungi datasets only have one training example per class, making LOOCV impossile to perform: the LOOCV accuracy is always 0. As a result, the Fungi penalty factor present in the csvs is the median of all tested factors.

The figures presented in the paper are generated with the exponent p=5.

How to compile and run

As a cmake project, most IDE should be able to open/compile/run the application. If using the command line (tested under Linux), use the following steps:

mkdir cmake-build-release
cd cmake-build-release
cmake -DCMAKE_BUILD_TYPE=Release ..
make ADTWNN1
./ADTWNN1

This last line run the application; it will complain that no argument are provided:

Error: argument parsing
ADTW NN1 classifier.
Usage:
  ./path/exec <ucr folder> <dataset name> <penalty> <threads>

The error message tells us what are the expected arguments:

• <ucr folder> path to a folder containing the UCR datsets, in the ts format.
• <dataset name> name of the dataset in the folder. The program will access the corresponding _TRAIN.ts and _TEST.ts files.
• <penalty> The additive penalty, should be tune per dataset
• <thread> Number of thread to use. Must be given, if only using 1 thread.

For example

./ADTWNN1 ~/DATASET/Univariate_ts Crop 0.5 8

Early Abandoned and Pruned Implementation (EAP)

The ADTW distance is efficiently implementated following

Early abandoning and pruning for elastic distances including dynamic time warping.

Data Min Knowl Disc (2021). https://doi.org/10.1007/s10618-021-00782-4

Herrmann, M., Webb, G.I.

Part of the code in this repository is borrowed from our Tempo project https://github.com/MonashTS/tempo.