- clone the repo
- cd to the root of the repo (i.e., the folder where
setup.pyis) - pip install with the -e option:
pip install -e .
- download the datasets.zip and the experiments.zip files
- unzip the
datasets.zipfile and paste its content underPortiloop>portiloop_software>dataset - unzip the
experiments.zipfile and paste its content underPortiloop>portiloop_software>experiments
The simulate_Portiloop_1_input_classification.ipynb notebook enables stimulating the Portiloop system and perform inference.
This notebook can be executed with jupyter notebook.
We enable easily using out trained artificial neural network on your own data to detect sleep spindles (note that the data must be collected in the same experimental setting as MODA for this to work, see our paper).
This is easily done by writing your signal in a simple text file, on the model of the example_data_not_annotated.txt file provided in the datasets.zip file.
Your file can then be directly used for inference in our offline_inference notebook.
Functions used for training are defined in python under the portiloop_software folder.
We provide bash scripts examples for SLURM to train the model on HPC systems.
Adapt these scripts to your configuration.
Your training curves can be visualized in real time easily using wandb (the code is ready, you may adapt it to your project name and entity).
The current hardware implementation (pynq FPGA with Vivado / Vivado HLS) is provided under the portiloop_hardware folder.