Example Code
Code and Documents
Attachment
Requirements:
- LabVIEW 2012
- G# framework (free download: http://www.addq.se/test-och-matsystem/g-framework-in-english/)
This library contains classes, functions and user interface for creating, training and using static feedforward neural networks. Sample project contains library, sample training data, VI that demonstrates training and User Interface VI for training networks using custom user training data.
In this version only static multilayer networks with tansig and linear output functions are supported.
Networks are objects made of network elements:
- Network inputs (processing and scaling input data)
- Network layers, subclasses:
*linear layer
*tansig layer
- Network outputs (processing and scaling output layer data)
Trainers are objects attached to network. Several trainer sublcasses are made, each implementing one of the following training methods:
- Single layer network training
- Gradient Descent backpropagation (GDBP)
- GDBP with momentum
- GDBP with adapting learn rate
- GDBP with momentum and adapting learn rate
- Levenberg-Marquardt backpropagation (single output and multiple outputs version)
Layers are initialized using Nguyen-Widrow layer initialization.
Networks can be saved and loaded using custom binary files (.lvnet) for later use in custom user VIs.
NN GUI.vi enables user to specify custom data training data from txt files, train network,track training process, plot training response and save/load networks to file.
At this time I am working on following extensions of code:
* radial basis layers/networks*
* time delay networks (dynamical networks)
* input data divide methods (for dividing training data into test and validation sets)
Please free to use code, leave feedback and contact me for any details you need.
Example code from the Example Code Exchange in the NI Community is licensed with the MIT license.
