// Initialize the experiment

v_init = -70
timestart = 1
numj = 7	// loop over time differences, odd number

forall {insert cldifus}
forall {Ra = 100}


// *******************************************************
// maximum segment length 5um and nseg = odd multiple of 5 (20 for dendr)
// *******************************************************

apical_dendrite[dendr] {nseg=int((int(L/5)+5)/40)*40+20}

for i=0,dendr_pre.size()-1 {
	apical_dendrite[dendr_pre.x[i]] {nseg=int((int(L/5)+5)/10)*10+5}
}
for i=0,dendr_post.size()-1 {
	apical_dendrite[dendr_post.x[i]] {nseg=int((int(L/5)+5)/10)*10+5}
}
for i=0,dendr_side.size()-1 {
	apical_dendrite[dendr_side.x[i]] {nseg=int((int(L/5)+5)/10)*10+5}
}

// *******************************************************

objref dendrv, posv, dendrv_side, posv_side, identv
dendrv = new Vector(0)
posv = new Vector(0)
dendrv_side = new Vector(0)
posv_side = new Vector(0)
identv = new Vector(0)
dendrv.append(dendr)
posv.append(synpos)
identv.append(0)

// *******************************************************
// The following lines define the recording sites at segmentation points along the dendrites which are closest to the distances defined by "sl".
// *******************************************************

access apical_dendrite[dendr]
orig = distance(synpos)
curr_pos = synpos
curr_dist = orig
curr_seg = 0
bc = 1
if (curr_seg<sl.size()) {
	if (curr_pos-sl.x[curr_seg]/L<0) {
		bc=0
	}
} else {
	bc=0
}

while (bc>0) {
curr_dist = distance(curr_pos-sl.x[curr_seg]/L)
curr_pos = (curr_dist-distance(0))/L
dendrv.append(dendr)
posv.append(curr_pos)
identv.append(-1)
dist_D = orig - sl_orig.sum(0,curr_seg) - curr_dist
curr_seg = curr_seg+1

if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-dist_D
	if (curr_pos-sl.x[curr_seg]/L<0) {
		bc=0
	}
} else {
	bc=0
}
}

if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-(curr_dist-distance(0)) 
}

for i=0,dendr_pre.size()-1 {
access apical_dendrite[dendr_pre.x[i]]
curr_dist = distance(1)
curr_pos = 1
bc = 1
if (curr_seg<sl.size()) {
	if (curr_pos-sl.x[curr_seg]/L<0) {
		bc=0
	}
} else {
	bc=0
}
while (bc>0) {
curr_dist = distance(curr_pos-sl.x[curr_seg]/L)
curr_pos = (curr_dist-distance(0))/L
dendrv.append(dendr_pre.x[i])
posv.append(curr_pos)
identv.append(-1)
dist_D = orig - sl_orig.sum(0,curr_seg) - curr_dist
curr_seg = curr_seg+1

if (i<dendr_side.size()) {
apical_dendrite[dendr_side.x[i]] {
	L_side = L
	orig_side = distance(0)}
if ((distance(1)-curr_dist)<L_side) {
	dendrv_side.append(dendr_side.x[i])
	posv_side.append((distance(1)-curr_dist)/L_side)
	ident_side.append(orig_side) 
	}
}
if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-dist_D
	if (curr_pos-sl.x[curr_seg]/L<0) {
		bc=0
	}
} else {
	bc=0
}
}

if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-(curr_dist-distance(0))  
}
} 

sl = sl_orig.c

access apical_dendrite[dendr]
orig = distance(synpos)
curr_pos = synpos
curr_dist = orig
curr_seg = 0
bc = 1
if (curr_seg<sl.size()) {
	if (curr_pos+sl.x[curr_seg]/L>1) {
		bc=0
	}
} else {
	bc=0
}
while (bc>0) {
curr_dist = distance(curr_pos+sl.x[curr_seg]/L)
curr_pos = (curr_dist-distance(0))/L
dendrv.append(dendr)
posv.append(curr_pos)
identv.append(1)
dist_D = orig + sl_orig.sum(0,curr_seg) - curr_dist
curr_seg = curr_seg+1

if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]+dist_D
	if (curr_pos+sl.x[curr_seg]/L>1) {
		bc=0
	}
} else {
	bc=0
}
}
if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-(distance(1)-curr_dist)  
}

for i=0,dendr_post.size()-1 {
access apical_dendrite[dendr_post.x[i]]
curr_dist = distance(0)
curr_pos = 0
bc = 1
if (curr_seg<sl.size()) {
	if (curr_pos+sl.x[curr_seg]/L>1) {
		bc=0
	}
} else {
	bc=0
}
while (bc>0) {
curr_dist = distance(curr_pos+sl.x[curr_seg]/L)
curr_pos = (curr_dist-distance(0))/L
dendrv.append(dendr_post.x[i])
posv.append(curr_pos)
identv.append(1)
dist_D = orig + sl_orig.sum(0,curr_seg) - curr_dist
curr_seg = curr_seg+1

if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]+dist_D	
	if (curr_pos+sl.x[curr_seg]/L>1) {
		bc=0
	}
} else {
	bc=0
}
}
if (curr_seg<sl.size()) {
	sl.x[curr_seg] = sl.x[curr_seg]-(distance(1)-curr_dist)  
}
}
 

dendrv.append(dendrv_side)
posv.append(posv_side)
identv.append(ident_side)

for i=0,dendrv.size()-1 {
print dendrv.x[i]
print posv.x[i]
}

// *******************************************************
// Run the experiment
// *******************************************************

nchan = dendrv.size()*2+2
print nchan

access soma

objref stimulator1, stimulator2, stimulator3, datamat, paramat 
objref vsoma
objref vdendr[(nchan-2)/2]
objref ca[(nchan-2)/2]
objref casoma
objref tvec
objref distv

datamat = new Matrix()
paramat = new Matrix(4,numi*numj*numk+1)
distv = new Vector()
vsoma = new Vector()
for i=0,(nchan-2)/2-1 {
	vdendr[i] = new Vector()
	ca[i] = new Vector()
}
casoma = new Vector()
tvec = new Vector()

objref syn1[numi], syn2[numi], syn3[numi]


finitialize(v_init)
fcurrent()
run()

soma{
stimulator1 = new IClamp(0.5)
stimulator1.del=stimstart
stimulator1.dur=1
stimulator1.amp=3
//stimulator2 = new IClamp(0.5)
//stimulator2.del=stimstart+5
//stimulator2.dur=1
//stimulator2.amp=3
//stimulator3 = new IClamp(0.5)
//stimulator3.del=stimstart+10
//stimulator3.dur=1
//stimulator3.amp=3
}

// loop over position
for ic=0,numi-1 {

if (ic>0) {
	syn1[ic-1].gmax = 0
	//syn2[ic-1].gmax = 0
	//syn3[ic-1].gmax = 0
}

	if (numi>1) {
		xvari = posv.x[ic]
	} else {
		xvari = synpos
	}
	
	// loop over conductance
	for kc=0,numk-1 {

	if (kc==0) {
		gi = 0
	} else {
		gi = gi_0 + gi_inc*(kc-1)
	}
	
	// loop over time
	for jc=0,numj-1 {

	count = ic*numk*numj+kc*numj+jc
	
	if (numj>1) {	
		tstart = stimstart-timestart+jc*0.5
	} else {
		tstart = stimstart
	}
	
	apical_dendrite[dendrv.x[ic]] {
	
	syn1[ic] = new gaba(xvari)
	syn1[ic].onset = tstart
	syn1[ic].tau = tau
	//syn1[ic].tau1 = tau1
	//syn1[ic].tau2 = tau2
	//syn1[ic].tau3 = tau3
	syn1[ic].gmax = gi
	syn1[ic].e = v_init
	
	//syn2[ic] = new gaba(xvari)
	//syn2[ic].onset = tstart+5
	//syn2[ic].tau1 = tau1
	//syn2[ic].tau2 = tau2
	//syn2[ic].tau3 = tau3
	//syn2[ic].gmax = gi
	//syn2[ic].e = v_init
		
	//syn3[ic] = new gaba(xvari)
	//syn3[ic].onset = tstart+10
	//syn3[ic].tau1 = tau1
	//syn3[ic].tau2 = tau2
	//syn3[ic].tau3 = tau3
	//syn3[ic].gmax = gi
	//syn3[ic].e = v_init

	}

	
	distv.resize(nchan)	

	access soma
	vsoma.record(&soma[0].v(0.5))
	casoma.record(&soma[0].cai(0.5))
	distv.x[0]=distance(0.5)
	
	for i=0,(nchan-2)/2-1 {
		vdendr[i].record(&apical_dendrite[dendrv.x[i]].v(posv.x[i]))
		ca[i].record(&apical_dendrite[dendrv.x[i]].cai(posv.x[i]))
		access apical_dendrite[dendrv.x[i]]
		distv.x[i+1]=distance(posv.x[i])
	}
	
	tvec.record(&t)
	
	finitialize(v_init)
	fcurrent()
	run()
	
	datamat.resize(vsoma.size()-(stimstart-timestart-1)/dt+1,nchan*numi*numj*numk+3)
	
	print (numk*numj*ic+numj*kc+jc+1)/(numi*numk*numj)

	datamat.setcol(nchan*(count),vsoma.remove(0,(stimstart-timestart-1)/dt))
	for i=0,(nchan-2)/2-1 {
	datamat.setcol(nchan*(count)+i+1,vdendr[i].remove(0,(stimstart-timestart-1)/dt))
	}
	datamat.setcol(nchan*(count)+(nchan-2)/2+1,casoma.remove(0,(stimstart-timestart-1)/dt))
	for i=0,(nchan-2)/2-1 {
	datamat.setcol(nchan*(count)+(nchan-2)/2+1+i+1,ca[i].remove(0,(stimstart-timestart-1)/dt))
	}	

	paramat.x[0][(ic*numk*numj+kc*numj+jc)] = stimstart
	access apical_dendrite[dendrv.x[ic]]
	paramat.x[1][(ic*numk*numj+kc*numj+jc)] = distance(xvari)
	paramat.x[2][(ic*numk*numj+kc*numj+jc)] = tstart
	paramat.x[3][(ic*numk*numj+kc*numj+jc)] = gi
	
	
	}
	}

	syn1[ic].gmax = 0
	//syn2[ic].gmax = 0
	//syn3[ic].gmax = 0
}


paramat.x[0][numi*numj*numk] = nchan
paramat.x[1][numi*numj*numk] = numi
paramat.x[2][numi*numj*numk] = numj
paramat.x[3][numi*numj*numk] = numk


datamat.setcol(nchan*numi*numj*numk,tvec.remove(0,(stimstart-timestart-1)/dt))
datamat.setcol(nchan*numi*numj*numk+1,distv)
datamat.setcol(nchan*numi*numj*numk+2,identv)

datamat.fprint(0,savdata,"%g;")
paramat.fprint(0,savparam,"%g;")

savdata.close()
savparam.close()