Qt Backpropagation Neural Network
=================================

Basic implementation of backpropagation neural network written in Qt for uni project.
Licenced under GPLv3.

The original design was done by Chesnokov Yuriy in his C++ implementation
over at CodeProject.com. The network can contain arbitrary number of layers
and each layer can have arbitrary number of neurons. However the implementation
uses only feed-forward each-to-each connection between the neurons.

The example UI is probably confusing and not intuitive at all, but it was done
only to demonstrate the network capabilities to the teacher,so take it as that.


How to use
==========

All the network logic is implemented in network.cpp and neuron.cpp, mainwindow.cpp
and race-client.cpp then contain examples of usage.

The code contains three example of setting and using the network. First one shows
number recognition 5x5 raster, where all the network initialization is done in
the code, including the training set (see MainWindow::recognizeNumber() method).

Second example features loading the network data from files (example file included).
The parsing is super-shitty, but...meh, good enough :)

The third example was an actual assignment, which was to train a race car, that can
follow supplied path. Basically we had to implement a race client that connects to
a race server, where bunch of neural-net-trained cars can race against each other.
Unfortunately I can't distribute the server nor the access to the public server,
so the client implementation will have to do.

Setting the network is as easy as

	m_network = new Network(QList<int>() << 25 << 18 << 8 << 4, 0.8f, 0.1f, 0.05f);

which will create a network with 4 layers (25 input neurons, 4 output), alpha = 0.8,
learning rate = 0.1 and max allowed network error is 0.05.

Training the network is also easy:

	m_network->trainSet(trainSet, outputs);

where trainSet is QList<QList<float> > (each 'inner' list is one training set)
and outputs is the expected outputs for the train set, same as the
trainSet - QList<QList<float> >. If the network was trained successfully, trainSet(..)
returns true, otherwise false.

Running the network on data is done by

	QList<float> *netOutputs = m_network->run(runset);

where runset is one set of data, ie. QList<float>. This returns list with network
outputs on the given set.

The code is hosted on github under GPLv3, so feel free to clone it, use it, share it,
improve it, just keep the copyrights in place and follow the licence.

https://github.com/mck182/QNeuralNet