Help Topics     Concepts     Package     Class


Decision Tree Bean Properties and Use

Properties

Mode
Select one of the following agent modes:
Train implies that the tree will be constructed from the data once all the data is read.
Test implies that the tree will classify and compute the accuracy of data as it is processed.
Run implies that the tree will classify data as it is processed.
Discretization interval
Data presented in numeric form is assigned to a discrete set of intervals. The discretization interval specifies the number of intervals used. The length of each interval is the difference between the maximum and minimum value encountered in training divided by the discretization interval.
Tree built
This label indicates whether a decision tree has been constructed.

Use

The Decision Tree Bean panel is used to create a classifier network. The Mode is set so that the network bean can be trained or used to provide an independant data source to test that training is sufficient.

Steps in using the panel for training include:

  1. Create and initialize an Import Bean and connect its data buffer to a Decision Tree Bean..
  2. Set the mode to train.
  3. Press the Initialize button.
  4. Create the tree by pressing the Cycle button on the Agent Editor toolbar.
  5. Create an inspector on the decisionTree parameter to see the tree created.

After training, use the panel for testing like this:

  1. Set the mode to test. Press the OK button to apply changes.
  2. Create inspectors on the predictClass parameter and the inputBuffer.
  3. Optionally connect another Import Bean if you wish to use independant data for testing.
  4. Press the Step button on the Agent Editor toolbar and view the input data and tree result in the inspectors.
  5. Create an inspector of the percentCorrect parameter to calculate how many test records are correctly classifed.

In run mode, data is processed similar to test but statistical properties like percentCorrect are not calculated.

If you press the Reset button, you reset the bean's statistics such as total records processed and percent correct. This is used to evaluate a tree's decisions from a different dataset. If you change data set definitions, or if new values are added outside the range calculated for the existing tree, press the Initialize button before training the tree. Initialize deletes any existing tree.