A working artificial brain
An Artificial Neural Network implemented from scratch based on the neural structure of the human brain using Python and the Numpy library. Features (set of input) from a training data file are sent to the input nodes of the neural network. The neural network is then trained over a number of epochs adjusting the weights for its connections based on the actual output through forward and back propagation. Through backpropagation the neural network figures out which nodes are to blame for the error in the network and adjust the weights of the connection between the nodes accordingly. Finally, after being trained the testing data is then used as input to the neural network and an output is spit back on a scale of 0 to 1.
Make sure you have the SciPY stack installed on your system.
Linux & OS X:
git clone https://github.com/AGontcharov/Artificial-Neural-Network.git
cd Artificial-Neural-Network
chmod u+x NeuralNetwork.pyWindows:
Not yet availableThe Neural Network can be tuned to achieve better (or worse) results by modifying some of the control variables located inside NeuralNetwork.py
By default these control variables have the following values:
- Learning Rate = 0.5
- Momentum = 0.01
- Maximum Error toleration = 0.00001
- Maximum number of epoch = 1000
The rest of the variables can be changed when the program is executed.
The Artificial Neural Network accepts a number of arguments that must be supplied to it and are listed in the following order:
| Argument | Description |
|---|---|
| arg1 | Path to the training data set |
| arg2 | Number of input nodes |
| arg3 | Number of hidden nodes |
| arg4 | Number of output nodes |
| arg5 | Path to the testing data set |
The number of elements (columns) in the feature must match the number of input nodes. Likewise, the number of elements in the output must match the number of output nodes.
Linux & OS X:
./NeuralNetwork.py [arg1] [arg2] [arg3] [arg4] [arg5]The output produced on the screen follow this format:
input node values, desired output node values, actual output node values
Alexander Gontcharov – [email protected]