Help Topics
Concepts
Package
Class
Self-Organizing Map Bean Properties
and Use
The Self-Organizing Map Bean panel provides these options:
- Architecture
- The network architecture consists of these parameters:
- Inputs, which is calculated when beans are
generated from the Training File.
- Output rows, which is the number of rows
in the network bean's output layer.
- 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.
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:
- Set the architecture input value to the number of outputs
from the bean providing data.
- Set the number of output rows and columns.
- Press the Set Architecture button.
- Train the network by pressing the Step, Cycle, or Run
buttons on the Agent Editor toolbar.
- 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.