T-Vectors

A method for generating person identifying vectors from arbitrary lengths of compatible EEG. See .

Please cite this article if you continue to explore and adapt the use of T-Vectors.

This work was created in large part using the DN3 library: https://github.com/SPOClab-ca/dn3. If confused about the implementation please refer to DN3 for more information.

Currently, the configuration files found in configs/ do not have any associated toplevel entries. This is because you will need to point the configuratron to wherever you have downloaded your data.

Once this is done, pretraining can be accomplished by running:

.../T-vectors/$ python3 pretrain_tvectors.py --median-sampling

T-Vectors can be extracted for each dataset ds, with T-Vector model weights t-weights.pt by running:

.../T-vectors/$ mkdir -p extracted_vectors/
.../T-vectors/$ python3 extract_vectors.py extracted_vectors/ds.npz t-weights.pt --session-id

If using provided pre-trained model weights (available for download in the Releases section of this repo), results and the figures found in the associated publication can be generated as follows:

.../T-vectors/$ ./make_predictions.sh
.../T-vectors/$ ./make_plots.sh

Each python module has a variety of options to change how training, extraction and analysis can be done, have a look by adding --help to any invocation.