Spatio-temporal Alignement

Guide to reproduce the results of the paper "Spatio-temporal alignements: Optimal transort through space and time" (https://arxiv.org/abs/1910.03860).

If your platform contains GPUs, please set the number of devices you would like to use in the beginning the scripts run_tsne....

1. Installation

The implementation of the proposed STA is available in the package sta provided in this folder. Before installing it, please make sure you have a miniconda environment installed and the following necessary dependencies (available through pip or conda):

• numpy
• cython
• joblib
• matplotlib
• scikit-learn
• soft-dtw (https://github.com/mblondel/soft-dtw/tree/master/sdtw)
• torch
• pandas
• numba
• pyts (https://pyts.readthedocs.io/en/latest/)

To reproduce the brain imaging experiment, you will also seed the MNE package (available with pip):

• mne (https://mne.tools/stable/index.html)

If you want 3D visualization of the brain signals, you also need

• mayavi
• pysurfer

Then proceed to the sta folder and run:

python setup.py develop

2. Experiments

2.1 theoretical_bound

• run plot_example.py to produce Figure 2 of the paper.
• run plot_bound.py to produce the theoretical bound (Figure 3)

2.2 Brain imaging

0. Make sur mne is installed.
1. Open run_tsne_brains.py to set the n_gpus and n_jobs params and run it to reproduce Figure 5.

2. To reproduce Figure 4, verify your installation of mayavi and pysurfer and run plot_brains.py.

   This last step can eventually take time because the MNE-Sample data must be downloaded.

2.3 Handwrittern letters

0. Run process_chars.py to generate and save the processed data.
1. Run plot_chars.py to visualize the chars (figure 6)
2. Open run_tsne_brains.py to set the n_gpus and n_jobs params and to compute and save the tsne maps
3. run plot_tsne_chars.py to reproduce Figure 7.