Welfare AI 2.0

This repository houses code for the welfare AI project done in collaboration with the McGill Animal Science Department.

Folders

Locomotion

This folder contains the excel files of the scaled coordinates passages.

Amanda_Notebooks

Models in these notebooks were trained by separating the data (25-75) BEFORE data augmentation.

Dylan_Notebooks

Models in these notebooks were trained by separating the data (25-75) AFTER data augmentation.

Results

This folder contains the final confusion matrices for the RF, CNN and LSTM trained by Amanda.

Running the code

Configuration

To run this code, you will need an Anaconda environment prepackaged with Python 3. This code was tested using python version 3.7.11. Python libraries including matplotlib, scikit-learn, numpy, keras, tensorflow and pandas must be installed in your python environment.

Python version

• python 3.7.11

Other libraries

You can install these libraries using conda install -c conda-forge --library_name.

• matplotlib 3.4.2
• scikit-learn 0.24.2
• numpy 1.19.2
• pandas 1.2.4
• keras-gpu 2.3.1
• tensorflow-gpu 2.1.0

Running this Code

Download all the files from the zip folder in this repository. Open a terminal and cd to the project folder (if you are using a python environment, make sure to activate it before). You can then open the jupyter notebook and run the cells.

git clone https://github.com/bioinfoUQAM/welfare_ai_2.0/

cd welfare_ai_2.0

python jupyter notebook

Then navigate to one of the jupyter notebooks using the Notebook Dashboard and click on it to open.

Before running the notebooks, make sure to add the correct paths to the scores.csv files and Locomotion directories:

# Get the list of locomotion data files 
listLocomotionData = os.listdir("../Locomotion")

# Initialize the data dictionary
data = {}
# Get the list of sensor names
sensor_names = pd.read_excel("../Locomotion/" + listLocomotionData[0]).columns