ERLANN - a neural network library for Erlang

======

What does this do?

• The library automates the creation of the neural network from a user defined number of input perceptrons, number of hidden layers and hidden perceptrons for each layer, and the number of output nodes.
• The library currenty accepts time-series data or a list of data as input data for the network.
• Data is preprocesssed.
• The library trains the network for 80% of the randomized set of training points and tests the network for the remaining 20% as testing points.
• The neural network is evaluatied using the MSPE which is also a stopping criteria for training.

Specifications

• Design: Single perceptron = single Erlang process
• Data Preprocessing: Log Normalization
• Activation Function: Sigmoid Function
• Learning Algorithm: Backpropagation
• Training Sequence: Training -> Testing -> Error Evaluation
• Evaluation Criteria: Mean Squared Prediction Error (MSPE)

References

• Erlang and Neural Networks by Wil Chung
• Fast Artificial Neural Network (FANN) library

License

GNU GPL v3

Copyright

2013 MAGNUM TEAM. Mindanao State University - Iligan Institute of Technology.