e-den syntax

WARNING: heavy-going reading ahead. Don't read this page unless you feel compelled by curiosity. You can probably answer all the queries you have about e-den syntax simply by playing with the Bug Lab ... And besides, you don't need to know any e-den syntax to play host to your share of the e-den master grid.

Still here? Don't blame me then.

The interpretation of each digit of the e-den genomes is at all times guided by two parameters: the gene mode and the gene sub-mode. There are 11 gene modes, one for each digit plus "Junk" mode (gene mode 10). (The major gene modes and the sub-modes applicable to each are explained in detail below). The zero gene mode is not being used, at present, and defaults to Junk mode. (In fact, earlier descriptions of the e-den syntax treated them as synonymous). In designing your organisms, however, this default action should not be relied upon as it is likely to change.
In most cases, the digit used to represent the gene mode is also used to set the mode. Thus, a skeletal gene always begins with a 9 (for "New Gene" or "Start") followed by a 1 (for "Skeletal" mode); a visual gene begins with a 9 followed by a 2 (for "Visual" mode), and so on. This syntax applies to gene modes 1 to 8 and will also apply to gene mode 0 when the syntax is extended. It does not apply to gene modes 9 and 10, however, because the 9 itself is reserved for marking the start and end of each gene and, as described below, gene mode 10, "Junk" mode, applies to the stretches of digits between genes.

Genes and Non-genes:Junk Mode

The Major Gene Modes

In the Bug Lab, the following convention is used beside each digit: a forward arrow indicates that choosing this digit will take you deeper into the syntactical tree or, at least, take you to a new window where you will asked to provide the necessary number of digits to assign to the value you have chosen; a backwards arrow indicates that choosing this digit will take you back to a more basic syntactical level or a previous screen; a colon (:) indicates that the sub-mode will revert immediately to the current one so that the screen will not change. If you have not already played with the Bug Lab to see the effect of choosing various digit combinations, it is recommended that you do so now. In fact, it is a good idea to print this page and have the Bug Lab open as you read it.

When a genome is being analysed, the initial gene mode is always "Junk". After the first 9 occurs, indicating the start of a new gene, the next non-nine digit sets one of the major gene modes as described in more detail below:

0 Reserved
1 Skeletal
2 Visual neurons
3 Metabolic
4 Change Current Neuron
5 Multiconnect
6 Alphabetical (Species identifier)
7 General neurons
8 Neurocopy
9 New Gene

Major Gene Modes... Top... Contents...

Major Gene Modes... Top... Contents...

0 Restore Tendency [2]
1 Threshold [2]
2 Drift [1]
3 Post-Fire-Change [2]
4 Sensitivity [2]
5 Neuron Number [2 in visual, 3 in general neuron]
6 Axon [7]
7 Flux [2]
8 Dendrite [7]
9 End Gene
(If any of these terms are unfamiliar, you should consult the e-den neurology page).

A Visual Gene such as ...925121294519... put through the decoding utility (decode.exe) might look as follows:

junk, junk, junk.....9

(visual)
2(start)
5(number)12
1(threshold)29
4(sensitivity)51
9(end)

junk,junk,junk....

This gene sets the "current visual neuron" to number 12, assigns the value 29 to that neuron's threshold, assigns the value 51 to that neuron's sensitivity and then ends. Any number of extra lines after the 2(start) and before the 9(end) could be inserted, provided they were of the form
x(sub-mode)nnn
where x was a digit from 0-8 and nnn was a string of digits of the appropriate length for that sub-mode.
But what if nnn were not the right length?
The syntax is blind to the designer's intentions and slavishly reads the expected number of digits, so that an attempt to assign a value of 5 to the sensitivy instead of 51 could NOT be implemented like this:

junk, junk, junk.....9

(visual)
2(start)
5(number)12
1(threshold)29
4(sensitivity)5 (WRONG, should be zero-five)
9(end)

junk,junk,junk....

The above sequence would instead be interpreted as:

junk, junk, junk.....9

(visual)
2(start)
5(number)12
1(threshold)29
4(sensitivity)59
x(previously part of the junk)nnn
...(no end in site)

Note that this means that single-digit neuron numbers (1,2,3 etc) must be referenced as 01,02,03 etc in Visual mode, or as 001, 002, 003 etc in General Neuron mode.

For most sub-modes, the digits supplied after the sub-mode is entered are interpreted literally as a 2 or 3 digit integer. For the drift parameter, the equation drift=2n-9 is employed, so that drift is assigned avalue ranging from 9, if n equals 9, to -9, if n equals 0. For the axon parameter, a seven-digit string is used to assign values to more than one parameter. The first digit is used to choose an axon number from the 10 available (0-9); the second digit is used to select an axon type (there are arbitrary associations between each digit and each type); the next three digits determine the recipient of the axon, the neuron that will receive its ouput (note that this is a three digit number for both visual and general neuron genes); the final two digits determine the effect of the axon, which corresponds to the strength of the output or the synaptic weight. A similar seven digit sequence is used for specifying dendrites.

WARNING: There is a potential trap here for the unwary. The zero-numbered axon and dendrite do not actually exist as axons and dendrites. They provide a method of allowing neurons to modify themselves and can be used to encode long term potentiation, for instance. Regardless of what recipient is specified for the zero-axon or dendrite, the software redirects the axon or dendrite to its own neuron. Although this conjurs up an image of an axon leaving its own neuron and looping back, this makes no biological sense and it would be better to think of these cennections as fictions standing in for some intrinsic cell process that self-adjusts the neuron. The software, however, treats the axon and dendrites as though they did loop back on to their source neuron.

The difference between Visual and General Neuron Mode

The 25 prototype visual neurons can be considered as a prototype visual segment. Each of these prototype neurons is actually a set of neurons, taken from all the visual stacks, such that each member of the set occupies the same segment position. This segment position is the same as the visual neuron number. Thus, in a one-eyed organism, prototype visual neuron #01 is the set of neurons #101, #126, #151 and #176. In a two-eyed organism, the same prototype is the set of neurons #101, #126, #151, #176, #201, #226, #251 and #276. If you examine the vision section on the neurology page, you will see that each these neurons receives information about the distance to the nearest one-atom (grass-atom). Thus, while neuron #101 is the neuron that receives this information from the east visual field of the first eye, prototype visual neuron #01 is the set of all grass-seeing neurons, including #101 and also those with neuron numbers 101+25n, for a range of values of n. Similarly, #02 is the set of all two-seeing neurons, #03 is the set of all three-seeing neurons and so on. The visual genes allow you to change all of these functionally similar neurons at once, rather than repeating genetic instructions for each of the neurons in the set.

By convention, when specifying neural connections for these prototype neurons, they are considered to be equivalent to the first segment of actual visual neurons, the real neurons #101-#125; these neurons process information from the east visual field of the organism's first eye. (For more details on the arrangment of neurons processing visual information, consult the e-den neurology page ). If (and only if) the organism has already specified that its skeleton contains an eye, assigning a value to visual neuron #12 automatically assigns the value to neuron #112. An appropriate value is also assigned to the similarly positioned neurons in the south, west and north facing neuronal segments, in this case #137 (112+25), #162 (112+2x25) and #187 (112+3x25). If a second eye has already been appended to the developing organism, the same procedure is applied to neurons #212,#237,#262,#287, and so on until the organism runs out of eyes or neurons.
For most neuronal variables, the same value is assigned for all the relevant neurons. For neuronal connections, however, a functionally equivalent connection is created instead. Thus, if the gene specifies that visual prototype neuron #12 should have an axon connected to motor neuron #002, then neuron #112 (east) is connected to #002 (east) but neuron #137 (south) is connected to #003 (south), neuron #162 (west) to #004 (west) and neuron #187 (north) to #005 (north). In each case, the connection links the visual neuron to a motor neuron that makes the organism move in the same direction as the visual field. If #112 had been connected to #003, the south motor neuron, that is 90 degrees clockwise from its own east facing field, then the same clockwise relationship would have been applied to the other neurons. Other recognised relationships are opposite and anticlockwise. In fact, when operating in visual mode, the directions of the motor neurons shouldn't be thought of as east, south west and north but as same, clockwise, opposite and anticlockwise.
A similar logic applies to connections between visual neurons: if #112 were connected to #137, this would be interpreted as same eye, same segment position, 90 degrees clockwise; if #112 were connected with #138, this would be interpreted as same eye, 90 degrees clockwise, one neuron further along in the segment . Connections to neurons not associated with one of the 4 cardinal compass directions are not subject to this process and in this case the same connection is created for all visual neurons. For instance, if visual prototype neuron #01 is connected to #001, the stay-still neuron, then all visual neurons occupying the first position in their respective segments will also be connected to #001.(The auditory neurons, being associated with 8 compass directions, are treated as individual neurons rather than being associated with a particular direction; as recipients of visual connections, they are treated like #001 rather than #002.)

Not every gene needs to specify which neuron is to receive the assigned values. The current neuron number and current visual neuron number retain their values across genes. One gene can set the current visual neuron number to #12 (....925129.....) and a later gene can assign a value to its threshold (.....921299.....).
Both the current neuron number and the current visual neuron number begin with a default value of zero and, in both Visual mode and General Neuron mode, assigning a value to the zero-numbered neuron causes all neuronal parameters to be overwritten by those associated with the zero-neuron, which is thus a master prototype. In this way, all visual neurons or all neurons can be assigned a default value for drift, threshold and so on, and only neurons that need to depart from this pattern need to have those parameters reset. Axonal and dendritic connections assigned to these master prototypes are ignored.
In both modes, an attempt to assign a parameter beyond the available numbers will assign the parameter to a randomly selected neuron.

Major Gene Modes... Top... Contents...

0 Not in use [0]
1 Metab-Div [3]
2 Metab-Child [3]
3 Substitution Frequency [2]
4 Insertion Frequency [2]
5 Deletion Frequency [2]
6 Crossover Frequency [2]
7 Jump Frequency [2]
8 Incubation Period[2]
9 End Gene

The 3-digit values assigned to Metab-Div and Metab-Child are first multiplied by 10 and thus these parameters can have values ranging from 0 to 9990. The other parameters have values ranging from 0 to 99. It is important to note that the first assignment to each of these parameters is interpreted literally but, thereafter, the assignment contributes to a rolling average that has a 50% contribution from the new value and 50% from the previous average.

Major Gene Modes... Top... Contents...

......9

(general neuron)
7(start)
5(neuron number)112
(continue with assigments)....

Although the above statement could mutate into statements about another neuron if any of the digits "112" mutated, neuronal genes that failed to specify a neuron number and simply relied on the previous value could not mutate in this way. And, when a mutation occurred, all assigment statements referring to that neuron would be re-assigned as a result; there would be little chance for evolution to keep some of the assignment statements with the original neuron and move others to a new neuron. It could happen, if the combination 975nnn occurred at the right point in the genome, but this is relatively unlikely.
In designing the e-den syntax to encourage neural evolution, I wanted to increase the chance that existing assigment statements and connections would be applied to new neurons as a result of mutation. By offering a whole gene mode that does little else but change the current neuron number in various ways, the syntax creates a higher chance that new neurons will be brought into play by mutation. In a random genome, each gene mode is equally likely to follow any section of Junk and thus, with the current syntax, about 1 in 8 genes in a random genome can be expected to change the current neuron. It is too early to say whether this syntactical trick will actually have the desired effect; certainly, with designed genomes, there is a natural tendency to restate the neuron number at the start of each neuronal gene, circumventing the intention of this gene mode. As more natural organisms arise, however, I expect this to change.

The actual operation of this syntax is largely self explanatory, with the following options being available:

0 Reduce current number by 100
1 Reduce current number by 25
2 Reduce current number by 1
3 Increase current number by 100
4 Increase current number by 25
5 Increase current number by 1
6 Copy Neighbour
7 Copy Source to Current
8 Copy Current to Target
9 End Gene

Genes in this mode are of the form 94x... and when x is 0-5 the current neuron number is changed as indicated and the sub-mode reverts to "On StandBy", offering the same choices again. When x is 6, a new syntactical level is reached and a single digit is read before returning to "On StandBy"; that digit causes neural parameters to be copied from the current neuron to one of its neighbours or vice versa, or to the "generic" neuron or vice versa (the generic neuron is number #000, the master prototype).
When x is 7 or 8, all neural parameters are copied either to or from the current neuron, using the source or target neurons previously specified in the Neurocopy gene mode. (The source and target neurons both default to neuron #001 at the start of the genome and are thereafter changed in the Neurocopy mode, retaining their value across genes). These 8 options may be repeated within the gene in any conceivable order until a 9 ends the gene.

Major Gene Modes... Top... Contents...

The syntax of this gene mode is relatively complex. Starting in the "On StandBy" sub-mode, the following options are offered (the number in square brackets represents the number of digits expected after the sub-mode is set):

0 Set recipient to current [0]
1 Range [6]
2 Segments [1]
3 Connection Rule [1]
4 Recipient/Interval [3]
5 Axon Number [1]
6 Axon Type [1]
7 Effect [2]
8 Apply Pattern [0]
9 End Gene

Sub-modes 1 to 7 are used to assign values to the various parameters required by the Multiconnect() function; in each case the necessary number of digits is read and then the sub-mode reverts to "On StandBy". There are 8 parameters required for the function: start, end, segments, connection rule, recipient, axon number, axon type and effect.

The start and end parameters are set by the six digits read in the "Range" sub-mode. These two parameters set the range of neurons that will be susceptible to the Multiconnect() function. The start parameter has a default initial value of #101, which represents the first visual neuron, and the end parameter has a default initial value equal to the maximum neuron number, currently #400. Thus, if these parameters are not adjusted, the Multiconnect() function applies to all neurons after the first stack of 100. (Note that these parameters are shared with the Neurocopy() function and can also be adjusted in the Neurocopy sub-mode; they retain their values across genes).

The segments parameter further limits the range of neurons susceptible to the Multiconnect() process. It refers to which segments of 25 neurons within each stack of 100 will have connections applied. If the parameter has a value of 1,2,3 or 4, then the 1st, 2nd 3rd or 4th segments respectively of each stack will be susceptible. If it has a value of 5, then the 1st and 2nd segments will be suceptible, if 6 then the 3rd and 4th segments, if 7 the 1st and 3rd, and, if 8, the 2nd and 4th. Values of 9 or 0 leave all segments within the start-end limits susceptible. The neurons indicated by the combination of the range and segments sub-modes are those that will develop new axons during application of the Multiconect process.

There are four possible connection rules, which are assigned with the genetic sequence 3n; the rule chosen depends whether n takes one of the values indicated in square brackets: FORWARDS [0,1,2], BACKWARDS [3,4,5], SPECIFIC [6,7,8] and EQUIVALENT [9].

The recipient parameter is a 3-digit number that is used to determine where the new axons are directed. In the case of the SPECIFIC connection rule, the recipient parameter indicates the number of the neuron to which all new axons will be directed. In the case of the FORWARDS or BACKWARDS rules, this parameter indicates the interval between the axonal source and axonal target; if the connection rule is FORWARDS and the interval is 111 then all neurons satisfying the range/segments restrictions will be connected to a neuron 111 positions further along in the neuronal array (provided such a neuron exists). If the connection rule is EQUIVALENT then some simple directional logic is applied, similar to that employed in the visual gene mode (see above). This software interprets the recipient parameter as a literal recipient number for the first neuron in the range, and then tries to find a functionally equivalent recipient for every other neuron in the range.

The final parameters needed by the Multiconnect() function are the axon number, axon type and axon effect for the new connections. To continue our previous example, the entire set of parameters might specify that for all neurons in the range #101-#400 that fall within the 1st and 3rd segments of their respective stacks, their fifth axon should be of type 1 (direct excitatory) with an effect of 25, and should be connected to a neuron 111 neurons further along in the array (start = 101, end = 400, segments = 7/ODD, rule = 1/FORWARDS, recipient/interval = 111, axon number = 5, axon type = 1/EXCITATORY, effect = 25).

Note that these parameters may be set in any order and need not be re-specified with each Multiconnect gene; they retain their values across genes.

Also, note that the Multiconnect() function is not actually called unless the digit 8 is read while in "On StandBy" mode. It is possible to end the gene (with a 9) without calling the function at all.

Finally, when a zero occurs in "On StandBy" mode, it has two effects: it assigns the value of the current neuron number to the recipient parameter and also sets the connection rule. No digits are read and the sub-mode remains "On StandBy".

Major Gene Modes... Top... Contents...

Major Gene Modes... Top... Contents...

Major Gene Modes... Top... Contents...

The syntax of this gene mode is complex and is best explored using the Bug Lab, which offers the following options:

The major parameters required by the Neurocopy function are the source, target, start and end parameters, with the start and end parameters being set by six digits set within the Range sub-mode and the other parameters set by 3 digits within their own sub-modes. Note that the start and end parameters are shared with the Multiconnect mode.

In addition, the function needs three Boolean (TRUE/FALSE) parameters: copy soma, copy axons and dendrites, and copy literally. These all default to TRUE or ON with each new Neurocopy gene and thereafter toggle with each occurrence of the relevant digit (0,1 or 2).

Finally, the function can be called with one of three copying rules: source-to-target, source-to-range, or copy-in-segments. Each occurrence of the relevant digit (6,7 or 8) immediately calls the function with that rule, using whatever values the necessary parameters hold at that time. A 9 both calls the function with the source-to-target rule and also ends the gene.

The function uses the supplied parameters as follows. Basically, it copies parameters from the source neuron to one or more target neurons. If copy-soma is ON, then the values of threshold, drift, flux, sensitivity, restore-tendency, and post-fire-change are copied (these are collectively know as the "soma" variables). If copy-axons-and-dendrites is ON, then the neural connections are copied, overwriting previous uses of the relevant axons and dendrites (but not deleting existing connections in the target neuron if none are specified for that axon or dendrite in the source neuron).

If copy-literal is ON, then the soma variables simply overwrite the previous values in the target neuron. If copy-literal is OFF, however, then the source values are averaged with the target values to create a neuron with characteristics intermediate between the two. For connection variables, the copy-literal parameter determines whether the source connections are copied blindly, so that the target neuron ends up connected to exactly the same neurons as the source neuron, or whether the target neuron is provided with a functionally equivalent connection, in the sense that was discussed under the Visual gene mode (see above).

The three copying rules determine which neurons are the target of the copying process. The source-to-target rule simply copies values from the source neuron to the single neuron specified by the target parameter. The source-to-range rule copies from the source neuron to every neuron in the range specified. The copy-in-segments rule copies the source neuron to the target neuron, adds 25 to both and the re-copies, continuing for as long as both source and target stay with range (the values of source and target revert to their original values once the function completes, however).

All of this might sound very complex but it is a simple matter to specify, for instance, that a certain neuron should have some desired threshold and then call copy-to-range to set the same threshold to a whole bank of neurons. For example:

.....9

(general neuron)
7(start)
5(number)101
1(threshold)29
9(end)

9999999999999

(neurocopy)
8(start)
3(source)101
5(range)102200
7(copy to range)
9(end)

9...........

This sequence specifies that neuron number 101 should have a threshold of 29 and then copies the same threshold to neurons 102 to 200. As it stands, it also copies axonal and dendritic connections, which might not be desirable, but if the sequence occurred early in the genome there might not be any connections at this point. (If it were not intended that connections be copied, a single digit 1, placed before the 3, 5 or 7 in the above Neurocopy gene, would toggle connection-copying off).

Even if this gene mode is not used by a single human programmer, it is hoped that, like the Multiconnect mode, it will allow e-den evolution to create groups of neurons with similar functional capabilities that can work cohesively together, rather than having to evolve each neuron independently.

Major Gene Modes... Top... Contents...

Major Gene Modes... Top... Contents...