Skip to content

Latest commit

 

History

History
33 lines (17 loc) · 1.16 KB

README.md

File metadata and controls

33 lines (17 loc) · 1.16 KB

Identifying user presses using Neural Networks built with PyTorch

The raw data for 5 people was recorded using Peratech sensors, and the presses were collected whilst users were playing Guitar Hero 3. The notebooks show the procedure to train three models: CNN, LSTM and CNN-LSTM. The inner workings of these networks and the training procedure can be found inside the folder peratouch.

How to setup:

Install suitable version of PyTorch using pip.

Then in the repository directory install the remaining requirements and local project peratouch:

pip install -r requirements.txt -e peratouch

Where to start:

Run script five_users_to_npz.py to convert raw data into a npz file:

python3 scripts/five_users_to_npz.py

Open first notebook 01_data_processing.ipynb and run all cells to extract trigger sections from data.

Training the networks

Run notebook 02_networks.ipynb to train all networks.

Results and figures

The results of training the networks are stored under results/ To produce the figures, run the script

python3 scripts/create_result_plots.py

and figures will be stored under figures/