Cluster Templates¶

General¶

The cluster-templates are used to define the basic structure of a network. This defined structure is comparable to a plant trellis. It is basically structures in hexagons, which are called hexagons.

step1

General structure of the file:

version: 1
settings:
    neuron_cooldown: <COOL_DOWN>
    refractory_time: <REFRACTORY_TIME>
    max_connection_distance: <MAX_DISTANCE>
    enable_reduction: <ENABLE_REDUCTION>

hexagons:
    X,Y,Z
    X,Y,Z
    ...
    X,Y,Z

axons:
    X,Y,Z -> X,Y,Z

inputs:
    <NAME>: X,Y,Z
    <NAME>: X,Y,Z (binary)
    <NAME>: X,Y,Z (match)

outputs:
    <NAME>: X,Y,Z
    <NAME>: X,Y,Z (float)

In the template the position, type and name of the hexagons has to be defined. The size of the input and output is defined by the given data when starting a training.

version¶

At the moment this is only a placeholder an the version: 1 is the only supported value at the moment. There are still too much changes to handle multiple versions currently.

settings¶

The entries within this block are all optional. If not listed, the default is used.

neuron_cooldown

It the value of how much the potential of a neuron is reduced from one cycle to the next. As default it is so high, that the remaining potential has no impact on the next cycle anymore.
refractory_time

Gives the number of cycles until a triggered neuron can be triggered again by the input. Within this time-perios, only the cooldown of the neuron is active. (Default = 1; Minimum = 1)
max_connection_distance

Maximum distance in number-of-hexagon, which a synapses can reach from the source- to the target-neuron. (Default = 1; Minimum = 1)
enable_reduction

Enable reduction-process to cleanup network. (Defaul: false)

position¶

X,Y,Z the x-, y- and z-coordinates of the hexagon. All hexagons must be connected by at least one side with another hexagon.

Info

It is theoretically possible to order them 3-dimensional by using different z-values, but this was never tested until now.

target¶

Inputs and outputs is also a X,Y,Z and has to be the same position, like the hexagon, where this input or output should be connected to.

name¶

<NAME> of this input and output for identification to be able to add input- and output-values.

binary input (optional setting)¶

For the case that the input-data have only value 0 and 1, a (binary) has to be added at the end of the input:

inputs:
    <NAME>: X,Y,Z (binary)

Otherwise the results for binary input become really bad. Even the input on the hexagon with this flag is not a binary input, all input-values greater than 0 are automatically handled as 1.

matchint input (optional setting)¶

Use the (match) to use the match-mode for inputs.

inputs:
    <NAME>: X,Y,Z (match)

When this is set, for each input, which is not explizitly learned by the hexgon, a new synapse is created.

axons (optional setting)¶

The axons-section is not required and allow to connect any hexagon within the cluster with another one.

output-type (optional setting)¶

Use the (bool), (float) or (int) to modify the output.

outputs:
    <NAME>: X,Y,Z (bool)
    <NAME>: X,Y,Z (float)
    <NAME>: X,Y,Z (int)

(bool): outputs can only have value 0 or 1 and nothing between
(float: 32 outputs internally are combined bitwise to a float-value
(int): 64 outputs internally are combined bitwise to a uin64-value

Info

(bool) and (int) are still converted to a float at the end, before writing into a dataset-file. Will be chanded in the future.

Simple example¶

The following is a minimal example for a cluster-template.

version: 1
settings:
    neuron_cooldown: 100000000000.0
    refractory_time: 1
    max_connection_distance: 1
    enable_reduction: false

hexagons:
    1,1,1
    3,1,1
    4,1,1

axons:
    1,1,1 -> 3,1,1

inputs:
    input_hexagon: 1,1,1

outputs:
    output_hexagon: 4,1,1

It defines 3 hexagons. It contains an input-hexagon with the name test_input and an output-hexagon with name test_output. Based on their position, they are all in a straight line, like in the image below:

step1