All the used datasets are available on Zenodo:
Králik, Matej. (2020). Curated list of HAR datasets (Version 1) [Data set]. Zenodo. http://doi.org/10.5281/zenodo.3831958
This repository contains the implementation of various Classical Machine Learning (CML) and Deep Learing (DL) methods for Human Activity recognition (HAR) classification. The code uses the SciPy and PyTorch frameworks.
The original thesis, containing the full description of the work, as well as the complete set of results, is available in this repository.
Structure of the code in the repository:
- experiments - helper scripts for smaller experiments
- models - model definitions and running code
- plots - some of the output data from experiments
- sanity - sanity checking intermediate models used for testing
- environment.yml defines a conda environment used to run the application
Results from the original work were produced using the respective models, hyperparameter combinations and datasets present in models/run.py.
Running the code requires downloading the datasets and adjusting the relevant configuration variables locally.