A simple project that analyzes the error and accuracy of a neural network based on properties such as activation functions, number of neurons, and momentum.
To run the project on your operating system make sure you already have python 3 installed and then install the following packages:
RUN pip install --upgrade pip
RUN pip install opencv-python-headless
RUN pip install matplotlib keras tensorflow
If you have a docker environment:
docker build -t nndl . # run the build phase
docker container run --rm -v .:/app nndl # to test changes
In the local or dockerize environment, project execution creates error graphs for any combination of learning rate, momentum, and number of neurons. These results are stored in the results/errors folder and any other information, such as parameters and accuracy, are stored as logs in events.log in the same location.
To change the generated results, change the parameters in the properties.ini file.