This application uses a genetic algorithm to search for an optimum neural network structure.
It can currently handle Dense (Fully Connected), Conv2D, Dropout, MaxPooling and ActivityRegularization layers and branching neural network structures. It uses the Keras library to build, train and validate.
The inspiration for this project was this blog post (https://medium.com/@harvitronix/lets-evolve-a-neural-network-with-a-genetic-algorithm-code-included-8809bece164), which describes the original code from which this is forked, although it turned out that I had to rewrite pretty much the entire thing to incorporate the Convolution and other useful layers...
To run the genetic algorithm:
python3 evolutionary_neural_network_generator.py
Options are viewable via the -h switch
Example usage for breeding MNIST classifiers using a population of 10 networks over 20 generations: (Note that if you don't already have MNIST installed it will be downloaded for you, which might not work if you are behind a proxy)
python3 evolutionary_neural_network_generator.py -p 10 -g 20 -d mnist
To Run the Unit Tests:
python3 tests.py
You can set your network parameter choices by editing network_layer_options.py.
You can choose whether to use the MNIST or CIFAR10 datasets. Simply set -dataset switch to either mnist or cifar10.
You can also use your own datasets by using the --training_data and --test_data switches and following the data formatting instructions in the get_dataset_from_file() method in train.py,
or the examples in the test_train() function in tests.py