Help Topics
Concepts Package
Class
- Metric parameter
- Represents the metric used for selecting the best attribute at each
node. It could be one of the following: gainRatio
(0), chiSquare (1), laplace (2), gini (3), relief (4), contextInfo (5),
intraInterDist (6). The metric can be changed between different
runs of the algorithm.
- Discretization parameter
- Represents the number of intervals used for to discretize continuous
attributes. The discretization parameter can be changed between
different runs of the algorithm.
- numAttributes
- Represents the number of attributes. This property is set when the
decision tree bean is created based on the data in the input file, and cannot be changed
later.
- numClasses
- Represents the possible classes for a particular data set. This property
is set when the decision tree bean is created based on the data in the input file,
and cannot be changed later.
- numRecords
- Represents the number of records in the input file. This property is set
when the decision tree bean is created based on the data in the input file, and
cannot be changed later.
- Mode
- Select one of the following agent modes:
Train implies that the data is read from the Import corresponding
to the input file and used to construct the decision tree model.
Test implies that the decision tree model is used to classify new data and
compute the accuracy of classification based on known labels of test
data.
Run implies that the model is used to classify new data as it is
process. No feedback about accuracy can be given in this case, as the
correct labels are not known.
The Decision Tree Bean panel is used to create a decision tree classifier with the specified
parameters. The Mode should be set only to test or run values once the
input data has been stored in the model.
After the data is stored, the mode is automatically set to test. It can be
changed manually to run. The metric and discretization parameters can also be manually changed.
Steps in using the decision tree panel include:
- Set the mode, metric and discretization parameters, if necessary. Press the OK
button to apply changes.
- Create inspectors on the inspectable properties (inputBuffer,
outputBuffer, accuracy, error etc.).
- Optionally connect another Import Bean if you wish to use independent data for testing.
- Press the Step button on the Agent
Editor toolbar and view the input data and decision tree result in
the inspectors, or
- Press the Cycle button on the AgentEditor to cycle through all
the examples in the test set.
- Inspect the desired properties (e.g. accuracy) and change metric or
discretization parameters if the
current value is not satisfactory.
In run mode, data is processed similar to test but statistical
properties like accuracy and error are not calculated.
If you press the Reset button, you reset the
bean's statistics such as total number of examples processed, number of
correctly classified examples, accuracy and error.