SEMI_SUPERVISED_STRESS_DETECTION

Semi-supervised learning for wearable-based momentary stress detection in the wild

---

🔗 Quick Links

• 📍 Overview
• 📂 Repository Structure
• 🧩 Modules
• 🚀 Getting Started

📍 Overview

Python (3.7) & Pytorch (1.7.0) implementation for paper: Semi-supervised learning for wearable-based momentary stress detection in the wild

@article{yu2023semi,
  title={Semi-supervised learning for wearable-based momentary stress detection in the wild},
  author={Yu, Han and Sano, Akane},
  journal={Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies},
  volume={7},
  number={2},
  pages={1--23},
  year={2023},
  publisher={ACM New York, NY, USA}

The semi_supervised_stress_detection project leverages semi-supervised learning to enhance stress detection accuracy and reduce labeling efforts. It offers a structured architecture for developing stress monitoring models, combining ECG and GSR signal analysis with deep learning techniques like autoencoders and pretrained models. Key components include configuration setup in configs.py, training orchestration in train.py, and task execution management in run_tasks.py. Through innovative training methods and feature extraction, this project provides a valuable tool for real-time stress analysis, aiding in understanding and managing stress levels effectively.

---

📂 Repository Structure

└── semi_supervised_stress_detection/
    ├── README.md
    └── src
        ├── configs.py
        ├── extract_embedding.py
        ├── model
        │   ├── __init__.py
        │   ├── cnn.py
        │   └── resnet.py
        ├── model_rl
        │   ├── __init__.py
        │   └── autoencoder.py
        ├── run_tasks.py
        ├── train.py
        └── utils
            └── dataset.py

---

🧩 Modules

src

File	Summary
configs.py	Summary: In configs.py, defines dataset & training params for supervised stress detection. Includes batch sizes, LR, epochs, save paths, and pretrained model locations. Crucial for setting up supervised training in the stress detection system.
train.py	Code Summary:**train.py orchestrates training tasks utilizing configurable settings from configs.py. Facilitates model training and evaluation within the repository's stress detection architecture. Executed procedures streamline model development.
run_tasks.py	Code Summary:**run_tasks.py orchestrates training tasks using Torch data loaders. Determines training mode based on configurations: supervised or semi-supervised. Loads data, initializes models, and conducts training accordingly.
extract_embedding.py	Code in src/ trains stress detection models using semi-supervised learning for improved real-time stress monitoring. It enhances accuracy and reduces labeling efforts.

src.utils

File	Summary
dataset.py	Code Summary:** A script in semi_supervised_stress_detection repo that trains semi-supervised learning models using autoencoder architecture for stress detection, facilitating robust feature extraction for stress analysis.

src.model

File	Summary
resnet.py	Code snippet: run_tasks.pySummary: Orchestrates task execution, coordinating ML model training, feature extraction, and performance evaluation within the repository's structured machine learning pipeline.
cnn.py	Code Summary:**The model_conv1d class combines ECG and GSR signal encoders, applies dropout and fully connected layers to classify stress levels. Pretrained weights are loaded based on the provided configuration.

src.model_rl

File	Summary
autoencoder.py	Code snippet in semi_supervised_stress_detection/ extracts embeddings using CNN and ResNet models to enhance semi-supervised stress detection. Key role: powering feature extraction for stress analysis in the parent repository.

---

🚀 Getting Started

Requirements

Ensure you have the following dependencies installed on your system:

• Python: version 3.7

⚙️ Installation

1. Clone the semi_supervised_stress_detection repository:

git clone https://github.com/comp-well-org/semi_supervised_stress_detection

2. Change to the project directory:

cd semi_supervised_stress_detection

3. Settings & Run:

Settings of augmentations, batch size, learning rates, etc. should be configured in src/configs.py file.

Please add your datasets in the src/utils/dataset.py file. For the labeled datasets, we aim to output the data format as Sequence(s): Channels (C) x Length (L) and Label; whereas we will have the sequences and augmentation views for the unlabeled set, as discussed in the paper.

Use src/run_tasks.py to run experiments.