home *** CD-ROM | disk | FTP | other *** search
- =============================================================================
- README file for the example files eight_jordan.xxx
- =============================================================================
-
-
- Description: Jordan network (partially recurrent network)
- ============ for the task to learn the shape of a lying figure '8'.
-
- The task of this partially recurrent network is to predict the shape
- of a lying figure '8'. The problem is described in detail in
-
- J.L. Elman: Finding Structure in Time. Cognitive Science, 14:179-211, 1990
-
- The two input units code the (x, y)-Position of the current point of
- the curve, the output units the (x', y')-Position of the next
- point. Usually 16 points (patterns) are used to approximate the shape
- of the figure 8, the central crossing point (0.5, 0.5) appearing
- twice, depending on which direction the stroke takes. The difficulty
- for the network arises from the input pattern of this central crossing
- point for which the network must predict two different successors
- (output patterns) depending on the previous point.
-
- See the user manual for a detailed description of Jordan networks and
- their usage.
-
-
- Network-Files: eight_jordan.net
- ==============
-
- This network file contains a trained jordan network for the task to
- predict the figure of a lying eight described above. The standard
- configuration file for this network is eight_jordan.cfg (one 2D display
- only).
-
-
- Pattern-Files: eight_016.pat
- ============== eight_160.pat
-
- The pattern files differ only in the number of patterns they contain,
- indicated in the name of the pattern file. The larger file consists of
- 10 concatenations of the smaller one
-
- Hints:
- ======
-
- The easiest way to create Jordan or Elman networks is with the BIGNET
- panel from the info panel. All network parameters can then be
- specified in a special Jordan or Elman network creation panel called
- with the respective button in the BIGNET panel.
-
- If you want to train your own Elman or Jordan network from scratch,
- note to set the proper initialization function and initialization
- parameters. In this example, we use the following values:
- 1.0, -1.0, 0.3, 1.0, 0.5 (5 parameters).
-
- Remember to set the update function to JE_Order or JE_Special,
- depending on your task (see the SNNS user manual for more details).
-
- You may choose between four different learning functions. They are
- given here with some values for the learning parameters for which the
- training is relatively fast
-
- 1st 2nd 3rd 4th 5th
- JE_BP (Backprop) 0.2
- JE_BP_Momentum 0.2 0.5
- JE_Quickprop 0.3 2.0 0.0001
- JE_Rprop 0.1 50.0
-
- The behaviour of this network can very nicely be visualized with the
- network analyzer tool which can be called from the info panel with the
- GUI button as ANALYZER. The proceed as follows:
- Press ON and LINE (so that both buttons are highlighted) from the
- buttons at the right.
- Press SETUP and choose X-Y graph from the network analyzer setup panel.
- Choose the following values for axis, min, max, unit, grid:
- x 0.0, 1.0, 11, 10
- y 0.0, 1.0, 12, 10
- This specifies the display area to be [0, 1] x [0, 1] and the outputs
- of neurons 11 and 12 for x and y (the output units of the jordan
- network).
- Choose m-test: 16 in this network analyzer setup panel to test 16
- patterns in a multiple inputs test sequence (You may also choose to test
- more or less input patterns.
-
- Finally, press the button M-TEST to test the trained network for the
- number of input patterns specified.
-
-
- =============================================================================
- End of README file
- =============================================================================
-