Machine learning (CS-433) class project, in cooperation with the Integrated Neurotechnologies Laboratory (INL) at EPFL
Welcome! This repository contains the code regarding class project 2 of the Machine Learning course (CS-433) at EPFL.
Feel free to contact any of us in case you have questions:
- Tomas Brants (tomas.brants@epfl.ch, @tbrants)
- Arthur Deleu (arthur.deleu@epfl.ch, @ArthurDeleu)
- Alicia Gomez Soria (alicia.soriagomez@epfl.ch, @aliciasoria)
The raw data, obtained from the INL is not in this GitHub repository, since this data is confidential. Hence, all data and labels needed to run the ML-models in Python, are provided as .npy files and a .csv file respectively.
- The folder 'Matlab codes' is informative and used to compute the BLA and IL power spectra, along with the PAC power spectra.
- A visualization of the data (upper 6 figures) and -more importantly- the extracted features (lower 3 figures) can be found in the folder 'Visualization data'.
- The folder 'Data' contains all the data (.npy-files and .csv-file) to run the Jupyter Notebooks.
- 'Feature arrays' are the input data for the 1-channel CNN.
- 'Labels' contains the .csv file with smoothed bar press data and will be binned in Python to create class labels.
- 'PAC_afterCNN.npy' contains the feature vector after running the 1-channel CNN on the PAC colormaps.
- 'input_MLP' contains the IL and BLA power spectra, which are feeded into the MLP together with 'PAC_afterCNN.npy' data.
- The folder 'Jupyter notebooks' contains all the notebooks relevant for the machine learning models.
- 'CNN_main.ipynb' is the notebook linked to the 1-channel CNN with input the PAC-features. It also contains the baseline model.
- 'MLP.ipynb' is the notebook linked to the MLP with input the CNN feature vector and the BLA and IL power spectra.
- The other notebooks were used to generate and save the data/features as .npy-files and don't need to be ran.
- Install Python version 3.7. and the Python libraries Pytorch, Numpy, sickit-learn and Pandas.
- Run 'run.py'