Help Topics
Concepts Package
Class
- k parameter
- Represents the number of neighbors the algorithm is looking at in order
to decide the class for a new example. Can be set between different runs
of the algorithm until an optimal value is found for a particular data
set.
- numAttributes
- Represents the number of attributes. This property is set when the knn
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 knn 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 knn 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 stored in the knn model.
Test implies that the knn model is used to classify new data and
compute the accuracy of classification based on known labels of test
data. The optimal k value for a particular data can be found in this mode,
based on the accuracy observed.
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 knn Bean panel is used to create a knn 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 k parameter can also be manually changed.
Steps in using the knn panel include:
- Set the mode and k, 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 knn 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 k if the
current value is not satisfactory. Any time when the end of the test file
is read, some of the properties of the bean are reset to their default
values (e.g. accuracy, number of correctly classified examples etc.) and
the algorithm starts over.
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.