Install the QuickRBF Package

After downloading the QuickRBF package, you need to build it in your system. On Unix systems, you only need to untar the source code package, then type `make' to build all the programs of package. On other systems, consult `Makefile' to build them. Also, we offer the pre-built windows binaries.

These programs are included in QuickRBF package.

  • quickrbf : the main program.
  • datatrans : convert data format from UCI or MS EXCEL to our format.
  • datascale : scale the data from -1 to 1.
  • centerselect : select the centers from training instances randomly.

Data Sets Preparation

The format of training and testing data file is:

    [label] [index1]:[value1] [index2]:[value2]...
[label] is the target value of the training data. For classification, it should be a positive integer which identifies a class (multi-class classification is supported). [index] is an integer starting from 1, and [value] is a real number. The labels in the testing data file are only used to calculate accuracy or error.

Here is an example data file.

We offer a simple tool, datatrans, in QuickRBF package to help you to convert the data file to the our format. You can use "Microsoft Excel" or other software to edit your data file, then save the data file to *.csv file, which use the "," to separate the feature values. In addition, the last value is the class feature value.

Here is an example Excel .csv data file.

The data format of Excel csv is almost the same with the UCI format, but it use the space as separator. Our datatrans program supports both formats of them.

The usage of datatrans program is

    datatrans [your data file]
Then, you will get the formatted file naming [your data file].out.

Scaling the data

Scaling your data properly is very important in data classification experiments. So we also offer datascale program in our package to help you to scale your data. The default range is from -1 to 1, and you may edit the source file to change the setting.

The usage of datascale program is

    datascale [data file]
Then, you will get the scaled file naming [data file].scale.

Here is an example data after scaling.

Selecting the Centers

How to select good centers for RBFN is still a hot research issue. In this package, we only offer a simple tool, centerselect, to select the centers from training instances randomly. We will try to find a better approach in the future. However, if your training data is not large (less than 10,000), you may consider to use the all training instances as centers.

The usage of centerselect program is

    centerselect [training data file] [number of centers]
Then, you will get the center data file naming [training data file].[number of centers].

Do the Experiments

After preparing the dat file, you can use quickrbf to start your experiment.

The usage of quickrbf program is

    quickrbf [center data file] [training data file] [testing data file]

Here is an example of satimage in Statlog data sets.

  • quickrbf satimage.scale satimage.scale.t

    Read 4435 data in training data file.
    Read 4435 data in center data file.
    Read 2000 data in testing data file.

    1. Calculation of bandwidths used 0.000 seconds
    2.1 Generating K matrix used 313.914 seconds
    2.2 Cholesky Decomposition used 62.336 seconds
    2. Calculation of weights used 376.250 seconds
    3. Testing use 6.000 seconds

    The accuracy is 0.9255
    Total use 382.25 seconds