Update May 2002 version 24
new manual, new features

The new version of DDLab includes many new features, functions, improvements and revisions, both major and minor, since the last official release in Feb 1999. Some of these are listed below, with links to illustrations. Changes not listed will be clear from the manual.
A number of logical, presentation and "crash" bugs have also been fixed (and not too may added - I hope).

For previous updates since March 96, click Feb 99 and Sept 97.

Some new features and changes

• graph tools
• meta-graph
• network architecture
• wiring
• rules
• attractor basins
• Derrida plot
• space-time patterns

Graph tools

• DDLab now includes powerful graph tools which have two distinct applications. Firstly to represent the CA or RBN network itself - the ``network graph''. Secondly to represent the ``meta-graph'' of the basin of attraction field. Click here for examples.

• The attractor meta-graph represents the probability of jumping between basins of attraction due to 1-bit perturbations to attractor states in the the basin of attraction field.

• The network graph represents the CA or RBN network itself. Links can be changed, added and cut within the graph, but this does not affect the underlying network, which can be changed in the "wiring graphic" and "wiring matrix" as before.

• Both the network graph and attractor meta-graph allow very flexible methods for rearranging and unraveling the graph:
  • Single nodes, connected fragments, or whole components, can be dragged with the mouse, with ``elastic band''edges, according to inputs and/or outputs.
  • The distance of fragment links from a node can be restricted, i.e. dragging the node + its immediate link nodes (step 1), the node + immediate links + their immediate links (step 2), etc.
  • Predefined 1d, 2d and 3d blocks can also be dragged.
  • Nodes and links can be re-scaled, and have alternative presentations.
  • Unreachable or hard to reach nodes can be identified and isolated.
  • Nodes with the fewest links can be automatically moved to the outer edges. This makes it easy to unravel a graph.
  • The pre-programmed graph layouts available are a circle of nodes, a spiral, or a layout in 1d, 2d or 3d.
  • The graph (or fragments) can be rotated, flipped, expanded, contracted, "shaken", and various other manipulations can be performed.
  • The graph layout can be saved/loaded.
  • An ``ant'' can be launched into the network that moves according to the link probabilities (as in a Markov chain) keeping a count of node hits.

• The data from which the graph is generated can be shown as a table, called the ``adjacency matrix'' for networks, and the ``jump-table'' for the meta-graph.

• The network architecture prompt has been redesigned with a number of new options:
• Show the network as a graph as described above. Click for examples of the same "scale free" n=50 network, in circle layout with some nodes dragged, and "unscrambled".
• A new wiring graphic option, the 1d circle layout, with the same options as the 1d wiring graphic. Click for examples: 1, 2, and 3.
• Extra options for the wiring graphic:
  • Set a "block" in 1d, 2d or 3d, where wiring (and rule) options apply just to the block.
  • Set random wiring (for the network or block) biased according to inputs k. If a power distribution of k was set, this also gives a power distribution of outputs.
  • The frequency of inputs k, outputs, or both, can be shown as a histogram.
  • "Kill" a node, cut all output wires and input wires, except one input to itself, then set rule 2 (i.e. no change).
  • Delete a node, i.e. reduce the network by one cell (for the 1d networks only).

• When setting mixed-k, there's a new option to set normal distribution, or a power law distribution with a given power exponent.

• Force the Z-parameter higher/lower.
• Set a chain-rule (or rules). These are maximally chaotic rules where the degree of pre-imaging approaches 1 with increasing system size, and G-density approaches 0.

• New mutation option, where a rule is chosen randomly from the set of chain-rules (see above).
• When showing the actor meta-graph, the basins of attraction can be shown within the meta-graph nodes, or the layout can simply follow the meta-graph node layout, so can be rearranged in any way, click here for an example (k=3 rule 81, n=10).
• A new option allows single basin mutants to be shown together on one screen, click here for an example.
• The RBN reverse algorithm: a new method has been introduced that re-orders the network to increase the wiring overlap between successive cells. This minimizes the growth of the partial pre-image stack, reducing memory load and speeding computation. The method is now the default, but can be suppressed. The re-ordering is entirely local within the algorithm. It should make no (or very little) difference to the presentation of attractor basins. The re-order is displayed below the top right window.
• Data on attractor basins can now include a complete list of states in each basin, including the number of pre-images and level away from the attractor.

• A new measure, the Derrida coefficient is introduced. The Derrida coefficient is the log of the tangent of the initial slope of the plot, and varies between about +1.3 and -1.2, where positive values indicate increasing chaos, negative values indicate increasing order. Click here for examples

• The samples of glider rules have been increased.
• New option for showing the fractal nature of space-time patterns, the return map in a 2d-phase plane.
• frequency bins: As well as showing "frozen" cells in space-time patterns, a new function displays the frequency of 1s in window of time-steps. Cells are shown in color according to preset "bins". The default (and max) is 10 equal bins, but the number of bins and bin boundaries can be revised.
• Network graph: Space-time patterns can be shown in network graph layout at the same time as normal layout.
  For example a 1d circle layout for 1d CA, or 2d triangular layout for 2d where k=6 or 7. Any arbitrary layout can be implimented by rearanging the network graph.
• Probabilistic networks: 2 types of randomness can be set for space-time patterns,
  ``update'' probability gives some asynchronicity.
  ``output'' probability introduces noise.
  The degrees of randomness can be set independently, and both ``update'' and ``output'' can be set at the same time.