| Prev | Next | Start of Chapter | End of Chapter | Contents | Index | (4 out of 7)

Fit Tester

The Fit Tester block can evaluate the performance of any neural network in NeurOn-Line. Attach one of its action links to a neural network that has been trained, and attach the other action link to a data set for the neural network (not necessarily the data set used for training the network). You cannot attach more than one neural network or data set to the block.

Configuring

To configure the Fit Tester, choose configure from its menu. It displays the configuration panel below.

Specify the Fit Metric. Choose an option depending on the type of network you are using and the type of problem you are solving:

If you are fitting a function, choose rms (functional approximation).
If you are solving a classification problem, choose fraction misclassified (classified).
If you are estimating a probability density function, choose probability error (density estimation by rho net).

To evaluate the Fit Tester, either pass it a control signal or choose evaluate from its menu. The block applies the neural network to the data set's input data, and stores the result in the data set as the prediction data. It then compares the data set's prediction data and target data.

When the Fit Tester finishes evaluating, it passes a control signal and a scalar value. The scalar value tells you how well the prediction data matches the target data. The Fit Tester computes that number differently depending on the option you chose in the configuration panel:

If you chose rms error, it passes a positive number. The closer the number is to 0, the better the network fits the function. To compute the number, the Fit Tester subtracts the target value from the predicted value to get the error, squares that error, and takes the square root of the mean of the errors over the entire training set.
If you chose fraction misclassified, it passes a number from 0 to 1, where 0 means all samples were classified correctly, and 1 means no samples were classified correctly. To compute the number, the Fit Tester figures the ratio of misclassified samples to total samples.
If you chose probability error, the Fit Tester figures the negative mean of the logarithm of the probability predictions of the Rho Network. The lower the number, the more accurately the Rho Network matches the probability distribution implied by the examples in the data set.

In addition to the calculation of the scalar error, the function of the Fit Tester is to fill up the predictions matrix in the data set; otherwise, the predictions are empty.

To view the predictions stored in the data set, use the data set's configuration panel. For more information, see "Data Set".

Example

In this example, a Fit Tester tests a Backpropagation network after it has been trained. The network is trained and tested a total of five times. A Path Display shows you the error number to let you know how well the network fits the data it is training on.

The figure also shows the plot of predicted values against target values. Notice that they lie on a 45 degree line, which means the predicted and target values are equal, i.e., they are a good fit.

See Also

For more information on how to use this block, see the pages below.

Click here for more information...
Basic Block Behavior

"Neural Network Blocks" Chapter

Data Set

Trainer

Train and Test

Five Fold CV

| Prev | Next | Start of Chapter | End of Chapter | Contents | Index | (4 out of 7)

Copyright © 1996, Gensym Corporation, Inc.