Rat classification

A program that takes EEG and EMG data to label the sleep state of rats. REQUIRES python 3.6+.

25 hours of data takes roughly 4 minutes to classify, depending on platform.

Paper

Paper available here

Dependencies

pip install --upgrade tensorflow
pip install keras
pip install numpy
pip install scipy
pip install matplotlib
pip install h5py

How it works

Start the program like so:

python rat_classifier.py *data_to_classify* *path_to_save_labels*
EX:
python rat_classifier.py rat8.mat labeled8.mat

data_to_classify must contain a 4000 x N array named EEGandEEG on the form

EEGandEMGColumn = [EEG;
                   EMG]

where EEG and EMG both are 2000 x 1 vectors.

The labels will be saved in the path_to_save_labels in a 3 x N array named Labels The columns of the array will be one hot encoded like this

LabelColumn = [WAKE;
              SWS;
              REM]
or
LabelColumn = [0;
               0;
               1]
means that the rat is in REM sleep

The program uses a voting system, fifteen models will vote with their predictions and the sleep state with the highest percentage in total will "win". This makes the program more likely to correctly classify unknown cases.

Models

15 models can be found in the folder "models". However, more models can be added by poking around in the base_classifier.py code. Also don't forget to add its number to the MODEL_NUMS global in rat_classifier.py More models should mean better classification accuracy

Performance

At the moment this generalized classifier performs as follows on these rats.

Rat1: 96.25 %
Rat2: 91.31 %
Rat3: 93.27 %
Rat4: 87.56 %
Rat5: 86.40 %

dekvall / rat_sleep_patterns