This code is an attempt to help reproduce results from the paper "Time Series Alignment with Global Invariances" (https://openreview.net/forum?id=JXCH5N4Ujy).

This code is Python3 code and relies on the following libraries:

tslearn>=0.3
numba
geoopt
torch
matplotlib
numpy
scipy
imageio

Dependencies can be installed via:

pip install -r requirements.txt

(maybe the torch dependency will raise an error and an alternative install method will be prompted).

Also, the base folder of this code (path to this README.md file) should be added to your Python path for dtw_gi package to work properly:

export PYTHONPATH=$PYTHONPATH:path/to/base/folder

For figures 1 to 5, notebooks are provided that generate the figures:

• Fig. 1
• Fig. 2
• Fig. 3 (note that this Figure reports timings and hence obtained results can be different from one execution environment to the other, however, trends should be similar, since they reflect theoretical complexity)
• Fig. 4
• Fig. 5
• Fig. 6