Project Parrot is an object-oriented implementation of a Neural Network, trainable by backpropagation. The network structure is defined by a tuple of parameters contained in the class DNA, which makes it conveniently combinable with evolutionary algorithms.
The network structure can be set up with very few intuitive commads:
The network can then be trained saved and exported. The main.java contains a full example of all the methods available to pull the data, customize the structure, train the network, save results and export and the network itself.
All you need is to import a csv file with the structure below and feed it to the network:
An implementation of one evolutionary algorithm is project HelloDarwin, which extends the Parrot framework, by making it mutable and shapeshifting.
These tools were built during a research project that aimed to find the most efficient network configuration to solve different kinds of problems, among which time series forecasting.