BSOM1: Bayesian Self-Organizing Map Ver. 1.0
Initially, BSOM reads data from a file specified by the applet parameter `input'.
See the source of the present page.
A data file must have rows each of which represents the coordinates of a data point.
See the contents of sample data files.
Moreover, data files must be under the directory where this HTML file exists.
If you want to change data, enter a new file name in `Input File' and press the return key.
If the new data have been read, the centroids and hyperparameters are initialized.
By pressing the `learn' button, the learning of BSOM is started.
In this mode, hyperparameters are controlled manually using the sliders.
By pressing the `init' button, the positions of the centroids are initialized randomly.
By pressing the 'auto' button, the auto-learning of BSOM is started.
In this mode, hyperparameters are tuned by a hyperparameter search algorithm.
By pressing the `init' button, the positions of the centroids are initialized randomly
and the hyperparameters are also set to initial values.
The auto-learning is suppressed while you grab a slider,
By pressing the 'density' button, estimated density for the data is calculated
and displayed using the gray scale.
In addition, the present values of hyperparameters are displayed.
In this mode, the other processes are suspended.
The suspended processes are restarted by making this mode off.
The resolution of density calculation is
specified by `dstep', which is the pixel size of a cell.
Small dstep requires large calculation times.
Change Unit Size
If you want to change the number of units, enter the number to `#unit' and press the return key.
This value must be larger than 1 and smaller than 100.
Initial setting after changing the unit size takes somewhat of time.
By pressing the `output' button, the present coordinates of the centroids are outputted.
In case of Netscape Navigator, the results are outputted to the java console,
if it is made usable in the `Options' menu.
In case of Internet Explorer 3, the results are outputted to the java log file,
if is is made usable in the option setting.
Appletviewer in JDK is most usable since the results are outputted to the standard output.
See the header of the source file Bsom1.java
Sample Data Files
Each sample data set is made by adding Gaussian noise to points on
a simple mathematical curve.
The number of data points (n) or the standard deviation of the noise (sd)
- Sine+Gaussian Noise(sd=1/10)
- Circle+Gaussian Noise(n=100)
(1)If you read this page via the internet, you can use the above data files on my disk.
However, you cannot use your data files on your disk,
since applets loaded via the internet generally are forbidden reading local files on user's disk.
If you want to make the applet read your data, you need to load the class files from my site
into your disk and call the applet from the local files. See BSOM Software Package.
(2)Netscape Navigator 3 on Windows has a bug that applet parameters are ignored
unless the document encoding is Latin-1.
Thus, you need to set the encoding to Latin-1.
(3)I strongly recommend using browsers with a just-in-time (JIT) compiler.
Calculation on the other browsers is rather slow.
(4)When the hyperparameters are not changed at the start in the auto-learning,
you should increase beta slowly using the slider for a while.
(5)The newest information is in my
`Neural Network Lab Page'.
Nov 23 1996 updated
Nov 20 1996 created