def det(m):
hm = h.Matrix(m.shape[0], m.shape[1])
for i in range(int(hm.nrow())):
hm.setcol(i, h.Vector(m[:, i]))
e = h.ref(0)
d = hm.det(e)
return d * 10.0**e[0]
def expm(m):
hm = h.Matrix(m.shape[0], m.shape[1])
for i in range(int(hm.nrow())):
hm.setcol(i, h.Vector(m[:, i]))
hm = hm.exp()
mo = numpy.matrix(m)
print m
for i in range(int(hm.nrow())):
for j in range(int(hm.nrow())):
mo[i, j] = hm.getval(i, j)
return mo
def run_single_volts(param_set, stim_data, ntimestep = 10000, dt = 0.02):
run_file = '../../neuron_genetic_alg/neuron_files/bbp/run_model_cori.hoc'
h.load_file(run_file)
total_params_num = len(param_set)
timestamps = np.array([dt for i in range(ntimestep)])
h.curr_stim = h.Vector().from_python(stim_data)
h.transvec = h.Vector(total_params_num, 1).from_python(param_set)
h.stimtime = h.Matrix(1, len(timestamps)).from_vector(h.Vector().from_python(timestamps))
h.ntimestep = ntimestep
h.runStim()
out = h.vecOut.to_python()
return np.array(out),np.cumsum(timestamps)
def neuron_run_model(param_set, stim_name_list):
h.load_file(run_file)
volts_list = []
for elem in stim_name_list:
curr_stim = h5py.File(stims_path, 'r')[elem][:]
total_params_num = len(param_set)
timestamps = np.array([dt for i in range(ntimestep)])
h.curr_stim = h.Vector().from_python(curr_stim)
h.transvec = h.Vector(total_params_num, 1).from_python(param_set)
h.stimtime = h.Matrix(1, len(timestamps)).from_vector(h.Vector().from_python(timestamps))
h.ntimestep = ntimestep
h.runStim()
out = h.vecOut.to_python()
volts_list.append(out)
return np.array(volts_list)
def _get_nrn_voltages(self):
"""The extracellular data (n_contacts x n_samples)."""
if len(self._nrn_voltages) > 0:
assert (self._nrn_voltages.size() == self.n_contacts *
self._nrn_n_samples)
# first reshape to a Neuron Matrix object
extmat = h.Matrix(self.n_contacts, self._nrn_n_samples)
extmat.from_vector(self._nrn_voltages)
# then unpack into 2D python list and return
return [
extmat.getrow(ii).to_python() for ii in range(extmat.nrow())
]
else:
raise RuntimeError('Simulation not yet run!')
def paranormal0(pt):
#print ("paranormal0 ", pt)
# return the normal direction from the inner paraboloid that contains pt
assert (pt[1] == 0.0)
# two dimensional problem due to assertion.
x = pt[0]
z = pt[2]
lam = 0.0
# three equations in three unknowns (x, z) on parboloid (x', z') is pt
# unknowns x, z, lam
# a*x^2 - c*z = 0
# x + lam*2*a*x - x' = 0
# z - lam*c - z' = 0
# can reduce to single cubic equation in x but might as well solve
# by newton method. Jacobian is
# [2*a*x , -c, 0 ]
# [1 + 2*a*lam, 0 , 2*a*x]
# [0 , 1 , -c ]
b = h.Vector(3)
m = h.Matrix(3, 3)
dx = h.Vector(3)
iter = 0
for i in range(100):
mset(m, x, z, lam)
bset(b, x, z, lam, pt[0], pt[2])
m.solv(b, dx)
#dx.printf()
x += dx.x[0]
z += dx.x[1]
lam += dx.x[2]
iter += 1
if dx.sumsq() < 1e-14:
break
assert (iter < 100)
global maxiter
maxiter = iter if iter > maxiter else maxiter
return normgrad((x, 0, z))
from neuron import h
m = h.Matrix(3, 4, 2)
m.setval(1, 2, 100)
m.setval(2, 2, 200)
def sparse_print(m):
j = h.ref(0)
for i in range(int(m.nrow())):
print(i, end=' ')
for jx in range(int(m.sprowlen(i))):
x = m.spgetrowval(i, jx, j)
print(" %d:%g" % (j[0], x))
print()
sparse_print(m)
def _build(self, cvode=None, include_celltypes='all'):
"""Assemble NEURON objects for calculating extracellular potentials.
The handler is set up to maintain a vector of membrane currents at at
every inner segment of every section of every cell on each CVODE
integration step. In addition, it records a time vector of sample
times.
Parameters
----------
cvode : instance of h.CVode
Multi order variable time step integration method.
include_celltypes : str
String to match against the cell type of each section. Defaults to
``'all'``: calculate extracellular potential generated by all
cells. To restrict this to include only pyramidal cells, use
``'Pyr'``. For basket cells, use ``'Basket'``. NB This argument is
currently not exposed in the API.
"""
secs_on_rank = h.allsec() # get all h.Sections known to this MPI rank
_validate_type(include_celltypes, str)
_check_option('include_celltypes', include_celltypes,
['all', 'Pyr', 'Basket'])
if include_celltypes.lower() != 'all':
secs_on_rank = [
s for s in secs_on_rank if include_celltypes in s.name()
]
segment_counts = [sec.nseg for sec in secs_on_rank]
n_total_segments = np.sum(segment_counts)
# pointers assigned to _ref_i_membrane_ at each EACH internal segment
self._nrn_imem_ptrvec = h.PtrVector(n_total_segments)
# placeholder into which pointer values are read on each sim time step
self._nrn_imem_vec = h.Vector(n_total_segments)
ptr_idx = 0
for sec in secs_on_rank:
for seg in sec: # section end points (0, 1) not included
# set Nth pointer to the net membrane current at this segment
self._nrn_imem_ptrvec.pset(ptr_idx,
sec(seg.x)._ref_i_membrane_)
ptr_idx += 1
if ptr_idx != n_total_segments:
raise RuntimeError(f'Expected {n_total_segments} imem pointers, '
f'got {ptr_idx}.')
# transfer resistances for each segment (keep in Neuron Matrix object)
self._nrn_r_transfer = h.Matrix(self.n_contacts, n_total_segments)
for row, pos in enumerate(self.array.positions):
if self.array.method is not None:
transfer_resistance = list()
for sec in secs_on_rank:
this_xfer_r = _transfer_resistance(
sec,
pos,
conductivity=self.array.conductivity,
method=self.array.method,
min_distance=self.array.min_distance)
transfer_resistance.extend(this_xfer_r)
self._nrn_r_transfer.setrow(row, h.Vector(transfer_resistance))
else:
# for testing, make a matrix of ones
self._nrn_r_transfer.setrow(row,
h.Vector(n_total_segments, 1.))
# record time for each array
self._nrn_times = h.Vector().record(h._ref_t)
# contributions of all segments on this rank to total calculated
# potential at electrode (_PC.allreduce called in _simulate_dipole)
# NB voltages of all contacts are initialised to 0 mV, i.e., the
# potential at time 0.0 ms is defined to be zero.
self._nrn_voltages = h.Vector(self.n_contacts, 0.)
# NB we must make a copy of the function reference, and keep it for
# later decoupling using extra_scatter_gather_remove
# (instead of a new function reference)
self._recording_callback = self._gather_nrn_voltages
# Nb extra_scatter_gather is called _after_ the solver takes a step,
# so the initial state is not recorded (initialised to zero above)
cvode.extra_scatter_gather(0, self._recording_callback)
from neuron import h
from math import cos, sin, fabs
h.load_file('nrngui.hoc')
cmat = h.Matrix(2,2,2).ident()
gmat = h.Matrix(2,2,2)
gmat.setval(0,1, -1)
gmat.setval(1,0, 1) # not needed unless experimenting with z2 below
y = h.Vector(2)
y0 = h.Vector(2)
b = h.Vector(2)
def callback():
#print 'callback t=', h.t
z1 = y.x[0]
z2 = -fabs(cos(z1)) # jacobian element is db[1]/dy[0]
z2 = 0
gmat.setval(1,0, z2)
b.x[1] = -sin(z1) + z1*z2
nlm = h.LinearMechanism(callback, cmat, gmat, y, y0, b)
dummy = h.Section()
trajec = h.Vector()
tvec = h.Vector()
trajec.record(y._ref_x[0])
tvec.record(h._ref_t)
def constructive_neuronal_geometry(source, n_soma_step, dx):
objects = []
source_is_import3d = False
# TODO: come up with a better way of checking type
if hasattr(source, 'sections'):
source_is_import3d = True
cell = source
# probably an Import3D type
num_contours = sum(sec.iscontour_ for sec in cell.sections)
if num_contours > 1:
raise Exception('more than one contour is not currently supported')
if num_contours == 1:
# setup the soma
# CTNG:soma
branches = []
parent_sec_name = []
for sec in cell.sections:
if sec.iscontour_:
soma_sec = sec.hname()
x, y, z = [
sec.raw.getrow(i).to_python() for i in xrange(3)
]
# compute the center of the contour based on uniformly spaced points around the perimeter
center_vec = sec.contourcenter(sec.raw.getrow(0),
sec.raw.getrow(1),
sec.raw.getrow(2))
x0, y0, z0 = [center_vec.x[i] for i in xrange(3)]
somax, somay, somaz = x0, y0, z0
xshifted = [xx - x0 for xx in x]
yshifted = [yy - y0 for yy in y]
# this is a hack to pretend everything is on the same z level
zshifted = [0] * len(x)
# locate the major and minor axis, adapted from import3d_gui.hoc
m = h.Matrix(3, 3)
for i, p in enumerate([xshifted, yshifted, zshifted]):
for j, q in enumerate([xshifted, yshifted, zshifted]):
if j < i: continue
v = numpy.dot(p, q)
m.setval(i, j, v)
m.setval(j, i, v)
# CTNG:majoraxis
tobj = m.symmeig(m)
# major axis is the one with largest eigenvalue
major = m.getcol(tobj.max_ind())
# minor is normal and in xy plane
minor = m.getcol(3 - tobj.min_ind() - tobj.max_ind())
#minor.x[2] = 0
minor.div(minor.mag())
x1 = x0
y1 = y0
x2 = x1 + major.x[0]
y2 = y1 + major.x[1]
xs_loop = x + [x[0]]
ys_loop = y + [y[0]]
# locate the extrema of the major axis CTNG:somaextrema
# this is defined by the furthest points on it that lie on the minor axis
pts = []
pts_sources = {}
for x3, y3 in zip(x, y):
x4, y4 = x3 + minor.x[0], y3 + minor.x[1]
pt = seg_line_intersection(x1,
y1,
x2,
y2,
x3,
y3,
x4,
y4,
clip=False)
if pt is not None:
pts.append(pt)
if pt not in pts_sources:
pts_sources[pt] = []
pts_sources[pt].append((x3, y3))
major_p1, major_p2 = extreme_pts(pts)
extreme1 = pts_sources[major_p1]
extreme2 = pts_sources[major_p2]
major_p1, major_p2 = numpy.array(major_p1), numpy.array(
major_p2)
del pts_sources
if len(extreme1) != 1 or len(extreme2) != 1:
raise Exception('multiple most extreme points')
extreme1 = extreme1[0]
extreme2 = extreme2[0]
major_length = linalg.norm(major_p1 - major_p2)
delta_x, delta_y = major_p2 - major_p1
delta_x /= n_soma_step
delta_y /= n_soma_step
f_pts = [major_p1]
f_diams = [0]
# CTNG:slicesoma
for i in xrange(1, n_soma_step):
x0, y0 = major_p1[0] + i * delta_x, major_p1[
1] + i * delta_y
# slice in dir of minor axis
x1, y1 = x0 + minor.x[0], y0 + minor.x[1]
pts = []
for i in xrange(len(x)):
pt = seg_line_intersection(xs_loop[i],
ys_loop[i],
xs_loop[i + 1],
ys_loop[i + 1],
x0,
y0,
x1,
y1,
clip=True)
if pt is not None: pts.append(pt)
p1, p2 = extreme_pts(pts)
p1, p2 = numpy.array(p1), numpy.array(p2)
cx, cy = (p1 + p2) / 2.
f_pts.append((cx, cy))
f_diams.append(linalg.norm(p1 - p2))
f_pts.append(major_p2)
f_diams.append(0)
for i in xrange(len(f_pts) - 1):
pt1x, pt1y = f_pts[i]
pt2x, pt2y = f_pts[i + 1]
diam1 = f_diams[i]
diam2 = f_diams[i + 1]
objects.append(
SkewCone(pt1x, pt1y, z0, diam1 * 0.5,
pt1x + delta_x, pt1y + delta_y, z0,
diam2 * 0.5, pt2x, pt2y, z0))
else:
parent_sec_name.append(sec.parentsec.hname())
branches.append(sec)
else:
h.define_shape()
soma_sec = None
branches = []
for sec in source:
branches.append(sec)
# this is ignored in this case, but needs to be same length
# so this way no extra memory except the pointer
parent_sec_name = branches
#####################################################################
#
# add the branches
#
#####################################################################
diam_corrections = {None: None}
while diam_corrections:
all_cones = []
pts_cones_db = {}
diam_db = {}
for branch, psec in zip(branches, parent_sec_name):
if source_is_import3d:
x, y, z = [branch.raw.getrow(i).to_python() for i in xrange(3)]
d = branch.d.to_python()
else:
x = [
h.x3d(i, sec=branch)
for i in xrange(int(h.n3d(sec=branch)))
]
y = [
h.y3d(i, sec=branch)
for i in xrange(int(h.n3d(sec=branch)))
]
z = [
h.z3d(i, sec=branch)
for i in xrange(int(h.n3d(sec=branch)))
]
d = [
h.diam3d(i, sec=branch)
for i in xrange(int(h.n3d(sec=branch)))
]
# make sure that all the ones that connect to the soma do in fact connect
# do this by connecting to local center axis
# CTNG:connectdends
if psec == soma_sec:
pt = (x[1], y[1], z[1])
cp = closest_pt(pt, f_pts, somaz)
# NEURON includes the wire point at the center; we want to connect
# to the closest place on the soma's axis instead with full diameter
x, y, z, d = [cp[0]] + [X for X in x[1:]], [cp[1]] + [
Y for Y in y[1:]
], [somaz] + [Z for Z in z[1:]], [d[1]] + [D for D in d[1:]]
for i in xrange(len(x) - 1):
d0, d1 = d[i:i + 2]
if (x[i] != x[i + 1] or y[i] != y[i + 1] or z[i] != z[i + 1]):
# short section check
#if linalg.norm((x[i + 1] - x[i], y[i + 1] - y[i], z[i + 1] - z[i])) < (d1 + d0) * 0.5:
# short_segs += 1
axisx, axisy, axisz, deltad = x[i + 1] - x[i], y[
i + 1] - y[i], z[i + 1] - z[i], d1 - d0
axislength = (axisx**2 + axisy**2 + axisz**2)**0.5
axisx /= axislength
axisy /= axislength
axisz /= axislength
deltad /= axislength
x0, y0, z0 = x[i], y[i], z[i]
x1, y1, z1 = x[i + 1], y[i + 1], z[i + 1]
if (x0, y0, z0) in diam_corrections:
d0 = diam_corrections[(x0, y0, z0)]
if (x1, y1, z1) in diam_corrections:
d1 = diam_corrections[(x1, y1, z1)]
if d0 != d1:
all_cones.append(
Cone(x0, y0, z0, d0 * 0.5, x1, y1, z1, d1 * 0.5))
else:
all_cones.append(
Cylinder(x0, y0, z0, x1, y1, z1, d1 * 0.5))
register(pts_cones_db, (x0, y0, z0), all_cones[-1])
register(pts_cones_db, (x1, y1, z1), all_cones[-1])
register(diam_db, (x0, y0, z0), d0)
register(diam_db, (x1, y1, z1), d1)
# at join, should always be the size of the biggest branch
# this is different behavior than NEURON, which continues the size of the
# first point away from the join to the join
diam_corrections = {}
for pt in diam_db:
vals = diam_db[pt]
if max(vals) != min(vals):
diam_corrections[pt] = max(vals)
cone_clip_db = {cone: [] for cone in all_cones}
join_counts = {
'2m': 0,
'2s': 0,
'3m': 0,
'3s': 0,
'4m': 0,
'4s': 0,
'0m': 0,
'0s': 0,
'1m': 0,
'1s': 0
}
for cone in all_cones:
x1, y1, z1, r1 = cone._x0, cone._y0, cone._z0, cone._r0
x2, y2, z2, r2 = cone._x1, cone._y1, cone._z1, cone._r1
pt1 = numpy.array([x1, y1, z1])
pt2 = numpy.array([x2, y2, z2])
axis = (pt2 - pt1) / linalg.norm(pt2 - pt1)
left_neighbors = list(pts_cones_db[(x1, y1, z1)])
right_neighbors = list(pts_cones_db[(x2, y2, z2)])
left_neighbors.remove(cone)
right_neighbors.remove(cone)
if not left_neighbors: left_neighbors = [None]
if not right_neighbors: right_neighbors = [None]
for neighbor_left, neighbor_right in itertools.product(
left_neighbors, right_neighbors):
clips = []
# process the join on the "left" (end 1)
if neighbor_left is not None:
# any joins are created on the left pass; the right pass will only do clippings
x0, y0, z0, r0 = neighbor_left._x0, neighbor_left._y0, neighbor_left._z0, neighbor_left._r0
if x0 == x1 and y0 == y1 and z0 == z1:
x0, y0, z0, r0 = neighbor_left._x1, neighbor_left._y1, neighbor_left._z1, neighbor_left._r1
pt0 = numpy.array([x0, y0, z0])
naxis = (pt1 - pt0) / linalg.norm(pt1 - pt0)
# no need to clip if the cones are perfectly aligned
if any(axis != naxis):
if r0 == r1 == r2:
# simplest join: two cylinders (no need for all that nastiness below)
sp = Sphere(x1, y1, z1, r1)
sp.set_clip([
Plane(x0, y0, z0, -naxis[0], -naxis[1], -naxis[2]),
Plane(x2, y2, z2, axis[0], axis[1], axis[2])
])
objects.append(sp)
else:
# is the turn sharp or not
# CTNG:joinangle
sharp_turn = numpy.dot(axis, naxis) < 0
# locate key vectors
plane_normal = numpy.cross(axis, naxis)
radial_vec = numpy.cross(plane_normal, axis)
nradial_vec = numpy.cross(plane_normal, naxis)
# normalize all of these
radial_vec /= linalg.norm(radial_vec)
nradial_vec /= linalg.norm(nradial_vec)
# count the corners that are inside the other cone (for both ways)
# CTNG:outsidecorners
my_corner_count = count_outside(
neighbor_left,
[pt1 + r1 * radial_vec, pt1 - r1 * radial_vec])
corner_count = my_corner_count + count_outside(
cone,
[pt1 + r1 * nradial_vec, pt1 - r1 * nradial_vec])
# if corner_count == 0, then probably all nan's from size 0 meeting size 0; ignore
# if is 1, probably parallel; no joins
# if corner_count not in (1, 2, 3, 4):
# print 'corner_count: ', corner_count, [pt1 + r1 * radial_vec, pt1 - r1 * radial_vec] + [pt1 + r1 * nradial_vec, pt1 - r1 * nradial_vec]
if corner_count == 2:
# CTNG:2outside
# add clipped sphere; same rule if sharp or mild turn
objects += join_outside(x0, y0, z0, r0, x1, y1, z1,
r1, x2, y2, z2, r2, dx)
elif corner_count == 3:
sp = Sphere(x1, y1, z1, r1)
if sharp_turn:
# CTNG:3outobtuse
if my_corner_count == 1:
sp.set_clip([
Plane(x1, y1, z1, -naxis[0], -naxis[1],
-naxis[2])
])
else:
sp.set_clip([
Plane(x1, y1, z1, axis[0], axis[1],
axis[2])
])
objects.append(sp)
else:
# CTNG:3outacute
objects += join_outside(
x0, y0, z0, r0, x1, y1, z1, r1, x2, y2, z2,
r2, dx)
if my_corner_count == 1:
objects.append(
tangent_sphere(neighbor_left, 1))
objects[-1].set_clip([
Plane(x2, y2, z2, naxis[0], naxis[1],
naxis[2])
])
else:
objects.append(tangent_sphere(cone, 0))
objects[-1].set_clip([
Plane(x0, y0, z0, -axis[0], -axis[1],
-axis[2])
])
elif corner_count == 4:
sp = Sphere(x1, y1, z1, r1)
if sharp_turn:
# CTNG:4outobtuse
# join with the portions of a sphere that are outside at least one of the planes
sp.set_clip([
Union([
Plane(x1, y1, z1, axis[0], axis[1],
axis[2]),
Plane(x1, y1, z1, -naxis[0], -naxis[1],
-naxis[2])
])
])
objects.append(sp)
else:
# CTNG:4outacute (+ 1 more)
# join with the portions of a sphere that are outside both planes
objects += join_outside(
x0, y0, z0, r0, x1, y1, z1, r1, x2, y2, z2,
r2, dx)
# AND clip the cone to not extend pass the union of the neighbor's plane and the neighbor
if r0 == r1:
neighbor_copy = Cylinder(
x0, y0, z0, x1, y1, z1, r0)
else:
neighbor_copy = Cone(
x0, y0, z0, r0, x1, y1, z1, r1)
clips.append(
Union([
Plane(x1, y1, z1, -naxis[0], -naxis[1],
-naxis[2]), neighbor_copy
]))
join_type = '%d%s' % (corner_count,
's' if sharp_turn else 'm')
join_counts[join_type] += 1
if neighbor_right is not None:
# any joins are created on the left pass; the right pass will only do clippings
x3, y3, z3, r3 = neighbor_right._x0, neighbor_right._y0, neighbor_right._z0, neighbor_right._r0
if x2 == x3 and y2 == y3 and z2 == z3:
x3, y3, z3, r3 = neighbor_right._x1, neighbor_right._y1, neighbor_right._z1, neighbor_right._r1
pt3 = numpy.array([x3, y3, z3])
naxis = (pt3 - pt2) / linalg.norm(pt3 - pt2)
# no need to clip if the cones are perfectly aligned
if any(axis != naxis):
# locate key vectors
plane_normal = numpy.cross(axis, naxis)
radial_vec = numpy.cross(plane_normal, axis)
radial_vec_norm = linalg.norm(radial_vec)
# we check again because sometimes there are roundoff errors that this catches
if radial_vec_norm:
# is the turn sharp or not
sharp_turn = numpy.dot(axis, naxis) < 0
nradial_vec = numpy.cross(plane_normal, naxis)
# normalize all of these
radial_vec /= radial_vec_norm
nradial_vec /= linalg.norm(nradial_vec)
# count the corners that are inside the other cone (for both ways)
my_corner_count = count_outside(
neighbor_right,
[pt2 + r2 * radial_vec, pt2 - r2 * radial_vec])
corner_count = my_corner_count + count_outside(
cone,
[pt2 + r2 * nradial_vec, pt2 - r2 * nradial_vec])
if corner_count == 2:
# no clipping; already joined
pass
elif corner_count == 3:
pass
elif corner_count == 4:
# CTNG:4outacute (+ 1 more)
# already joined; just clip (only in mild turn case)
if not sharp_turn:
if r2 == r3:
neighbor_copy = Cylinder(
x2, y2, z2, x3, y3, z3, r3)
else:
neighbor_copy = Cone(
x2, y2, z2, r2, x3, y3, z3, r3)
#print 'cc=4: (%g, %g, %g; %g) (%g, %g, %g; %g) (%g, %g, %g; %g) ' % (x1, y1, z1, r1, x2, y2, z2, r2, x3, y3, z3, r3)
clips.append(
Union([
Plane(x2, y2, z2, naxis[0], naxis[1],
naxis[2]), neighbor_copy
]))
if clips:
cone_clip_db[cone].append(Intersection(clips))
#print 'join_counts:'
#print join_counts
for cone in all_cones:
clip = cone_clip_db[cone]
if clip:
cone.set_clip([Union(clip)])
#####################################################################
#
# add the clipped objects to the list
#
#####################################################################
objects += all_cones
return objects
