Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: A recurrent neural network solution

This repository supports Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: A recurrent neural network solution.

The final models and data supporting the published manuscript are archived here.

Contents

Train_LSTM.ipynb is a notebook that generates the model from the archived data.

Test_LSTM.ipynb is a notebook that shows you how to use the trained LSTM to predict GRFs from your own accelerometer data.

LSTM_Example.ipynb is a notebook that provides a tutorial of how a Long Short-Term Memory Network (LSTM) can be used to predict ground reaction force (GRF) data from accelerometer data during running.

pre_processing.py contains helper functions used in LSTM_Example.ipynb and Test_LSTM.ipynb.

data/ Contains example accelerometer data, GRF data, condition/demographic data, and LSTM model file. Supports Test_LSTM.ipynb and LSTM_Example.ipynb.

If you're going to train an LSTM model using Google Colab (recommended), make sure you utilize their GPU Runtime Type. You will need to adjust the path to data/ depending on how files are uploaded in Google Colab.

Questions?

Open an issue if you have a question or if something is broken. You can also email me at the address listed in the associated publication.