Server for Dreemcare

Description

The key features of this project are:

• Manage EDF Files: Upload an EDF File, anonymize it, and make it available for download. During the upload,the evolution of the first EEG channel is predicted with a LSTM Neural Network.
• Get Prediction Scores: Display the performance of the Neural Network using the Root Mean Square Error (RMSE) of the train and test sets.
• Train Model: Train the LSTM Neural Network with the EDF of your choice.

Requirements

• python
• virtualenv

Installation (through virtualenv)

To set up python 3 virtualenv, install the project dependencies and run the Django app. Enter the following into a terminal:

git clone https://github.com/Max-Pol/assignement.git
cd assignement/project_dreemcare
virtualenv -p /usr/bin/python3 venv
source venv/bin/activate
pip install -r requirements.txt

python manage.py migrate

python manage.py runserver

Script

Train_model

python manage.py train_model -f <file_id> [-n <nb_epoch>]

This script allows you to train the LSTM Neural Network with the EDF file of your choice. The new LSTM Neural Network model replaces the previous one in media/model/, and will be used for the next file upload.

The Trainscore and Testscore (RMSE) are also automatically saved and updated, as you can see in the Get Prediction Scores section.

Note: The <nb_epoch> (number of epoch) is set to 100 by default.

Notes

Anonymization & scores calculation are done during the upload as requested ("à la volée"). However, given the dataset size it can take a long time to compute the predictions.

For this reason, I truncate the dataset in edf_app.views, for my poor computer to calculate the RMSE during the upload. If you have a stronger computing power, just comment the lines concerned. However, when you run the script train_model.py, the RMSE are computed on the whole dataset.