Help Topics
Concepts Package
Class
- 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.
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:
- Create and initialize an Import Bean and connect its data
buffer to a Decision Tree Bean..
- Set the mode to train.
- Press the Initialize button.
- Create the tree by pressing the Cycle
button on the Agent Editor toolbar.
- Create an inspector on the decisionTree parameter to see
the tree created.
After training, use the panel for testing like this:
- Set the mode to test. Press the OK
button to apply changes.
- Create inspectors on the predictClass parameter and the
inputBuffer.
- Optionally connect another Import Bean if you wish to use
independant data for testing.
- Press the Step button on the Agent
Editor toolbar and view the input data and tree result in
the inspectors.
- 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.