dEARpression

This project contains the code corresponding to the paper dEARpression (tbd).

Project Structure

earables_derpression
├── README.md
├── poetry.toml
├── pyproject.toml
└── src
    ├── Algorithms
    ├── Datasets
    ├── utils
    ├── feature_extraction.py
    ├── ml_pipeline.py
    └── shap_values.py

The folder src contains the source code. As the code is based on the tpcp package, it is organized in this way:

• Datasets contains the different wrappers for the different data modalities (voice and IMU).
• Algorithms contain preprocessing (speaker diarization) and feature extraction steps (gait and motion as well as openSMILE and wav2vec for voice).
• utils contain utility functions and helpers.
• feature_extraction.py contains the pipeline to extract voice, gait, and motion feature from the raw data (wav-files and IMU-files).
• ml_pipeline.py contains the machine learning pipeline, which takes the extracted features and performs classification (based on BioPsyKit).
• shap_values.py contains a pipeline, which calculates SHAP-values based on BioPsyKit-pipelines.

Troubleshooting

Version Resolving

As both EarGait and BioPsyKit are based on different versions of tpcp, two different virtual environments are recommended to use this project. The feature should be extracted using the one with EarGait, the machine learning should be done with BioPsyKit.