/*
Genetic Algorithm
to find neurons capable of performing specified computations
Version 3
Evolution of the parameters of a local, recursive model of dendritic
developemnt as proposed by Samsonovich and Ascoli, Hippocampus (2004)
Selection for neuronal morphologies performing well in a
computation (linear summation, spike order detection,...)
by Klaus M. Stiefel, CNL, Salk Institute, 1-2006
Thanks to Michael Hines, Yale
Reference:
Mapping Function onto Neuronal Morphology.
Klaus M. Stiefel and Terrence J. Sejnowski
Journal of Neurophysiology
Death to fascism!
*/
// GA PARAMETERS
// ------------------------------------
allseed = 1
cellnr = 2 // original 16
generations = 1 // original 400
pmutated = .5
preplaced = .8
cutoff = .8
wtboth = 0.1
wtsize = 0.1
wtequal = 0.02 // relative weights for lineartest
config = 99 // synpase space type
// ------------------------------------
load_file("parameters.hoc") // load this to override default parameters
objref cell[cellnr], genome[cellnr]
objref selector, rank, score, scorecopy, scoresorted, sizes
rank = new Vector(cellnr)
score = new Vector(cellnr)
scoresorted = new Vector(cellnr)
selector = new Random() // behold! I shall decide the fate of your genome!
// obsolete load_file("stdgui.hoc")
load_file("nrngui.hoc")
load_file("stimulator.hoc")
load_file("neuromorph.hoc")
load_file("thegenome.hoc")
load_file("setsynapses.hoc")
// load_file("init.hoc")
// nrnmainmenu()
// MAKE INITIAL POPULATION
// ------------------------------------
makestimulators(2)
sizes = new Vector(cellnr)
print "Initial population"
for cells = 0, cellnr-1 {
// print "Cell: ", cells
genome[cells] = new thegenome(cells+allseed)
cell[cells] = new neuronmorph(cells)
cell[cells].synapsespace()
setsynapses(cells)
define_shape()
cell[cells].synapseinsert()
sizes.x[cells] = cell[cells].nall
}
load_file("xover.hoc")
load_file("lineartest.hoc")
// load_file("ordertest.hoc")
//load_file("display.hoc")
xopen("display.hoc")
load_file("savegeneration.hoc")
// RUN EVOLUTIONARY LOOP
// ------------------------------------
for generation = 0, generations-1 {
display()
print " "
print "Generation F", generation
rank = new Vector(cellnr)
score = new Vector(cellnr, 1)
// ordertest(sizes.mean()) // this runs the electrophysiological simulations and scores cells on the task
// lineartest(sizes.mean())
lineartest(sizes.mean())
for az=0, cellnr-1 { cell[az].synapseremove() }
objref cell[cellnr]
scoresorted = score.c
scoresorted.sort()
for az=0, cellnr-1 { rank.x[az] = scoresorted.indwhere("==", score.x[az]) }
scoremin = score.min()
scoremean = score.mean()
scorestdev = score.stdev()
savegeneration(generation)
objref cell[cellnr]
for cells = 0, cellnr - 1 {
// select genomes
therank = cellnr
while (therank > cellnr-1) { therank = int(abs(selector.normal(0, cellnr*cutoff))) }
genome[cells].newgene = genome[rank.indwhere("==", therank)].gene.c
// mutate genomes
if ( selector.uniform(0, 1) < .7*pmutated ) { genome[cells].pointmutate() }
if ( selector.uniform(0, 1) < .1*pmutated ) { genome[cells].dubmutate() }
if ( selector.uniform(0, 1) < .1*pmutated ) { genome[cells].delmutate() }
if ( selector.uniform(0, 1) < .05*pmutated ){ xover(cells, selector.discunif(1, cellnr-1)) }
// make a new generation
// print "Cell: ", cells
if (cells == 0) {genome[cells].newgene = genome[rank.indwhere("==", 0)].gene.c } // elitism
genome[cells].gene = genome[cells].newgene.c
cell[cells] = new neuronmorph(cells)
cell[cells].synapsespace()
setsynapses(cells)
define_shape()
cell[cells].synapseinsert()
}
}
// display()