| Prev
| Next | Start of Chapter | End of Chapter | Contents | Index | (2 out of 7)
Introduction
NeurOn-Line comes with several blocks that test and train neural networks.
You will find the Training Blocks palette under the Neural Networks submenu of
the Palettes menu:
Usually, you will follow this procedure to find the best neural network and train
it:
- Find the best neural network configuration.
Use either the Train and Test Block or the Five Fold CV block with several different neural networks to find which network solves your problem best. You can test different types of neural networks and different configurations of the same neural network.
You can also create your own testing algorithms with the Trainer and Fit Tester blocks.
- Train the neural network.
After you find the best neural network configuration, use the Trainer block with that network and all available data to train the network.
You might even add a step after the last step. If you are not sure whether all the
input values you have actually affect the output, use a Sensitivity Tester.
Removing unimportant inputs and repeating the training cycle may improve the
performance of your final network.
Basic Training and Testing
These blocks perform basic training and testing:
Finding the Best Network Configuration
These blocks help you find the best neural network configuration for your
problem:
Finding Which Inputs are Significant
The Sensitivity Tester block determines which inputs of a neural network you
actually need to predict the output. It accomplishes this by testing the influence of
each input on each output by using a trained neural network and a data set of
sample data.
| Prev
| Next | Start of Chapter | End of Chapter | Contents | Index | (2 out of 7)
Copyright © 1996, Gensym Corporation, Inc.
