Examples of org.encog.util.normalize.output.zaxis.OutputFieldZAxis

org.encog.util.normalize.output.zaxis.OutputFieldZAxis

Both the multiplicative and z-axis normalization types allow a group of outputs to be adjusted so that the "vector length" is 1. Both go about it in different ways. Certain types of neural networks require a vector length of 1. Z-Axis normalization is usually a better choice than multiplicative. However, multiplicative can perform better than Z-Axis when all of the values are near zero most of the time. This can cause the "synthetic value" that z-axis uses to dominate and skew the answer. Z-Axis gets its name from 3D computer graphics, where there is a Z-Axis extending from the plane created by the X and Y axes. It has nothing to do with z-scores or the z-transform of signal theory. To implement Z-Axis normalization a scaling factor must be created to multiply each of the inputs against. Additionally, a synthetic field must be added. It is very important that this synthetic field be added to any z-axis group that you might use. The synthetic field is represented by the OutputFieldZAxisSynthetic class. @author jheaton

    norm.setReport(new NullStatusReportable());
    norm.setTarget(target);
    norm.addInputField(a = new InputFieldArray2D(false,SAMPLE1,0));
    norm.addInputField(b = new InputFieldArray2D(false,SAMPLE1,1));
    norm.addInputField(c = new InputFieldArray2D(false,SAMPLE1,2));
    norm.addOutputField(new OutputFieldZAxis(group,a));
    norm.addOutputField(new OutputFieldZAxis(group,b));
    norm.addOutputField(new OutputFieldZAxis(group,c));
    norm.addOutputField(new OutputFieldZAxisSynthetic(group));
    return norm;
  }

Examples of org.encog.util.normalize.output.zaxis.OutputFieldZAxis

Related Classes of org.encog.util.normalize.output.zaxis.OutputFieldZAxis