This repository is for the paper: "MetaWearS: A Shortcut in Wearable System Lifecycles with Only a Few Shots" The source code can be found in the "src" directory. In particular, src/few_shot_train.py represents the training, finetuning and evaluation procedure in MetaWearS.

How to use this code

Dataset Preparation

1. Download Datasets: Download the desired wearable datasets (e.g., UCI HAR, WISDM, etc.) and place them in a designated directory (e.g., data/).
2. Preprocessing: Each dataset might require specific preprocessing steps (e.g., normalization, windowing). Refer to the dataset's documentation or existing scripts for guidance.
3. Data Format: Ensure the data is formatted appropriately for the few_shot_train.py script. Typically, this involves organizing the data into training, validation, and testing sets, with each set containing sensor readings and corresponding labels.
4. Configuration: Update the configuration file (e.g., config.yaml) with the correct paths to your preprocessed datasets.

Running few_shot_train.py

1. Dependencies: Ensure you have all the necessary Python packages installed. You can typically install them using pip install -r requirements.txt.

2. Configuration: Modify the config.yaml file to specify the desired parameters for training, finetuning, and evaluation. This includes:

   • Dataset paths
   • Model architecture
   • Hyperparameters (learning rate, batch size, etc.)
   • Few-shot settings (number of shots, number of ways)
   • Evaluation metrics

3. Execution: Run the few_shot_train.py script from the command line:

   python src/few_shot_train.py --config config.yaml

   • Replace config.yaml with the path to your configuration file if it's different.

4. Output: The script will output the training, finetuning and evaluation results to the console and save them in the specified output directory.