Help Topics     Concepts     Package     Class


Self-Organizing Map Bean Properties and Use

Properties

The Self-Organizing Map Bean panel provides these options:

Architecture
The network architecture consists of these parameters:
  1. Inputs, which is calculated when beans are generated from the Training File.
  2. Output rows, which is the number of rows in the network bean's output layer.
  3. Output columns, which is the number of columns in the network bean's output layer.
Learn Rate
Enter a value to control how much the network weights are changed during a weight update. Larger values cause more change. Learn rate is a real value between 0.0 and 10.0, with a typical starting value of 0.2.
Number of Passes
The value for the number of passes. Training the network will stop when attained.
Conscience
Enter a value to control the bias factor that determines the distance a losing unit can reach in order to win a competition. Conscience is a real value between 0.0 and 1.0.
Agent Mode
Select one of the following agent modes:
Train implies that the network bean's weights are unlocked, and network weights will be adjusted as data is processed.
Test implies that the network bean's weights are locked, and that error calculations will be performed as data is processed.
Run implies that the network bean's weights are locked and no error calculations are made.
Epoch (batch) updates
Select batch updates if network weights are updated only after a complete training epoch. Otherwise weights are updated after each record.

Use

The Self Organizing Map Bean panel is used to create a network with specified architecture and training parameters. 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:

  1. Set the architecture input value to the number of outputs from the bean providing data.
  2. Set the number of output rows and columns.
  3. Press the Set Architecture button.
  4. Train the network by pressing the Step, Cycle, or Run buttons on the Agent Editor toolbar.
  5. You may wish to press the Stop toolbar button, change the Learn Rate or Conscience, and start again. If you change the network architecture, press Set Architecture for the changes to take effect. Press the Reset Weights button to re-initialize the network weights before starting training again if you wish.

The output from the bean is the cluster index. When in test mode, the prototypeInput property will display the weights for the cluster winner. The confidence parameter will compute a measure of the cumulative distance between the input values and the cluster prototype value; a smaller distance indicates a closer fit between the input and the cluster prototype. To see all the prototype values in any mode, inspect the network weights array and set the number of columns equal to the number of network inputs. Each row will correspond to a cluster prototype.