Update Sept 1997 - version 21 - new features

Some of the new features added since the last release in March 96 are listed below. An updated manual covering all DDLab functions, with examples and illustrations, is in the process of being written.

• Increased neighbourhood range, k.
  Maximum k has been increased to 13 (from 9). In 2d, k=13 gives a diamond shaped neighbourhood. k can vary from 1-13 and effective k=0 can also be set.
• The k mix.
  For networks with mixed k, the proportion of different k can be specified, then set at random.
• Sequential updating.
  Updating can be sequential as well as parallel. The sequence order can be left to right, right to left, random, or set to some particular order. The order sequence can be saved to a file. The neutral order components in the space of all possible orders can be calculated, and presented graphically as quasi sub-trees. Sequential updating applies to both attractor basins and space-time patters. Space-time patterns can be toggled on the fly between parallel and sequential.
• Hypercube.
  The network wiring can be set to correspond to a n-dimensional hypercube if k=log₂(n), or log₂(n)+1 to allow for self wiring. The connections in the hypercube can be set according to a given probability.
• Canalizing.
  New options are provided for setting canalizing inputs, C, including different C for subsets of nodes with different k. The graphics for depicting canalization is extended to show the frequency of different degrees of canalization across the network.
• Derrida plot.
  Options for the Derrida plot are extended to restrict the max Hamming distance, i.e. to zoom into the lower left hand corner of the graph.
• Probabilistic networks
  For forward dynamics only, the value of a network element can update according to its rule with a given probability, equivalent to adding noise to the system.
• Samples of rule-space sorted according to rule class.
  Files are included with large samples of k=5, 6 and 7 rules that have been sorted according to the standard deviation of input entropy. A scatter plot of density against standard deviation discriminates between ordered, complex and chaotic rules, where complex rules feature large scale interacting structures such as gliders and domain boundaries. The dynamics of particular rules in the sample can be examined, for example by selecting from the plot with the mouse. New samples can be assembled and sorted automatically, thus providing an unlimited source of complex rules for further study.
• Data on attractors for large networks.
  For networks that are too large to generate attractor basins, data can be gathered by generating a histogram of the frequency of reaching different attractors, giving the number and relative size of basins, attractor length and average run in length.
• Data on "skeletons" for large networks.
  A skeleton in a RBN is a subset of network elements that has "frozen" or stabilized along a transient before reaching the attractor, according to some set criteria. Other elements in the network may still be changing. A histogram of the frequency of reaching different skeletons may be generated, giving the number and relative importance of skeletons, and average run in length.
  
• Mutation
  The mutation of successive of networks when drawing attractor basins can now be turned off.
• Rotating attractor basins and changing the default "fan" angle.
  The orientation of attractor basins or sub-trees can be altered, i.e. they may be rotated by some angle. The pre-image "fan" angle may be changed relative to the default. Both of these adjustments can be preset, or made during a repositioning "pause" between each successive basin.