Although activations can be propagated through the network without
patterns defined, learning can be performed only with patterns
present. A set of patterns belonging to the same task is called a
pattern set. Normally there are two dedicated pattern sets when
dealing with a neural network. One for training the network (training
pattern set), and one for testing purposes to see what the network has
learned (test pattern set). In SNNS both of these (and more) can be
kept in the simulator at the same time. They are loaded with the file
browser (see chapter 
). The pattern set loaded last is
made the current pattern set. All actions performed with the simulator
refer only to, and affect only the current pattern set. To switch
between pattern sets press the button 
in the control panel
(see figure on page ). It
opens up a list of loaded pattern sets from which a new one can be
selected. The name of the current pattern set is displayed to the
right of the button. The name equals the name body of the loaded
pattern file. If no pattern set is loaded, ``Patternfile ?'' is given
as indication that no associated pattern file is defined.
Loaded pattern sets can be removed from main memory with the 
button in the control panel. Just like the 
button it opens a
list of loaded pattern sets, from which any set can be deleted. When a
pattern set is deleted, the corresponding memory is freed, and again
available for other uses. This is especially important with larger
pattern sets, where memory might get scarce.