hxxp://qwiknet.home.comcast.net/~qwiknet
QwikNet offers a wide array of features that make designing, training and testing neural networks easy and fun. Design Features - Up to 5 hidden layers
- Unlimited number of neurons in each layer
- Automatic data scaling
- Unlimited training data file sizes
- On-line help is always available
Training Features - Choice of several advanced training algorithms
- Automatic prevention of neuron saturation
- Cross-Validation training
- Visually monitor the learning process and the network weights
- Choice of sigmoid, tanh, Gaussian, or linear activation functions
- Input noise and weight decay
Testing Features - Graphically view the network outputs and errors for both training and testing data files
- Create three-dimensional contour plots
- Error histogram plots show distribution of errors
- All plots can be created and updated during training
Export Features - Ability to output trained networks in several formats including;
- C/C++,
- Pascal,
- Matlab
|