def test_G_R_C_grating_response(nt, nr, dt, dr, A_g, a_g, B_g, b_g, delay_g,
A_rc, a_rc, B_rc, b_rc, delay_rc, w_rc, a_rg,
delay_rg, w_rg, w_id, k_id, orient, C):
network = pylgn.Network()
integrator = network.create_integrator(nt=nt, nr=nr, dt=dt, dr=dr)
w_g = integrator.temporal_angular_freqs[w_id].rescale(1. / pq.ms)
k_g = integrator.spatial_angular_freqs[k_id]
kx_g = k_g * np.cos(orient.rescale(pq.rad))
ky_g = k_g * np.sin(orient.rescale(pq.rad))
t, y, x = integrator.meshgrid()
Wg_r = spl.create_dog_ft(A=A_g, a=a_g, B=B_g, b=b_g)
Wg_t = tpl.create_delta_ft(delay=delay_g)
Krg_r = spl.create_gauss_ft(a=a_rg)
Krg_t = tpl.create_delta_ft(delay=delay_rg)
Krc_r = spl.create_dog_ft(A=A_rc, a=a_rc, B=B_rc, b=b_rc)
Krc_t = tpl.create_delta_ft(delay=delay_rc)
Kcr_r = spl.create_delta_ft()
Kcr_t = tpl.create_delta_ft()
ganglion = network.create_ganglion_cell(kernel=(Wg_r, Wg_t))
relay = network.create_relay_cell()
cortical = network.create_cortical_cell()
network.connect(ganglion, relay, (Krg_r, Krg_t), weight=w_rg)
network.connect(cortical, relay, (Krc_r, Krc_t), weight=w_rc)
network.connect(relay, cortical, (Kcr_r, Kcr_t), weight=1)
stimulus = pylgn.stimulus.create_fullfield_grating_ft(angular_freq=w_g,
wavenumber=k_g,
orient=orient,
contrast=C)
network.set_stimulus(stimulus)
network.compute_response(ganglion)
network.compute_response(relay)
network.compute_response(cortical)
Wg = ganglion.evaluate_irf_ft(w=w_g, kx=kx_g, ky=ky_g)
Wr = relay.evaluate_irf_ft(w=w_g, kx=kx_g, ky=ky_g)
Wc = cortical.evaluate_irf_ft(w=w_g, kx=kx_g, ky=ky_g)
Rg = C * abs(Wg) * np.cos(kx_g * x + ky_g * y - w_g * t +
np.angle(Wg)) / pq.s
Rr = C * abs(Wr) * np.cos(kx_g * x + ky_g * y - w_g * t +
np.angle(Wr)) / pq.s
Rc = C * abs(Wc) * np.cos(kx_g * x + ky_g * y - w_g * t +
np.angle(Wc)) / pq.s
assert (abs(Rg - ganglion.response) < complex(1e-12, 1e-12)).all()
assert (abs(Rr - relay.response) < complex(1e-12, 1e-12)).all()
assert (abs(Rc.clip(min=0 / pq.s) - cortical.response) < complex(
1e-12, 1e-12)).all()
def test_G_patch_grating_response(nt, nr, dt, dr, A_g, a_g, B_g, b_g, delay_g,
w_id, k_id, orient, C, patch_diameter, n):
network = pylgn.Network()
integrator = network.create_integrator(nt=nt, nr=nr, dt=dt, dr=dr)
w_g = integrator.temporal_angular_freqs[w_id].rescale(1. / pq.ms)
k_g = integrator.spatial_angular_freqs[k_id]
Wg_r = spl.create_dog_ft(A=A_g, a=a_g, B=B_g, b=b_g)
Wg_t = tpl.create_delta_ft(delay=delay_g)
ganglion = network.create_ganglion_cell(kernel=(Wg_r, Wg_t))
stimulus = pylgn.stimulus.create_patch_grating_ft(
angular_freq=w_g,
wavenumber=k_g,
orient=orient,
contrast=C,
patch_diameter=patch_diameter)
network.set_stimulus(stimulus)
network.compute_response(ganglion)
Rg = C * (X(a_g / patch_diameter, a_g * k_g, n) -
B_g / A_g * X(b_g / patch_diameter, b_g * k_g, n)) / pq.s
assert abs(Rg - ganglion.center_response[0]) < complex(1e-10, 1e-10)
def test_dog_patch_grating_response():
response_e = np.array([[0.1017374087, 0.2867197943, 0.5105320011, 0.1666709735],
[0.3492268988, 0.3616052645, 0.4750891407, 0.1640082272],
[0.4789040160, 0.4328530438, 0.3805594923, 0.1561941696],
[0.2884185736, 0.3175258969, 0.2569232750, 0.1437369651],
[0.1140384287, 0.1163952049, 0.1390511885, 0.1274369599],
[0.0339383730, 0.0179229458, 0.0526124870, 0.1083213337],
[0.0074631404, 0.0145706277, 0.0061438229, 0.0875612286],
[0.0013737748, 0.0074470223, -0.0077840845, 0.0663789388]]) / pq.s
k_max_id = 40
step = 5
patch_diameter = np.array([3, 1.5, 0.85, 0.3]) * pq.deg
response = np.zeros([int(k_max_id/step), len(patch_diameter)]) / pq.s
network = pylgn.Network()
integrator = network.create_integrator(nt=1, nr=7, dt=1*pq.ms, dr=0.1*pq.deg)
spatial_angular_freqs = integrator.spatial_angular_freqs[:k_max_id][::step]
Wg_t = tpl.create_delta_ft()
Wg_r = spl.create_dog_ft(A=1, a=0.3*pq.deg, B=0.9, b=0.6*pq.deg)
ganglion = network.create_ganglion_cell(kernel=(Wg_r, Wg_t))
for j, d in enumerate(patch_diameter):
for i, k_d in enumerate(spatial_angular_freqs):
stimulus = pylgn.stimulus.create_patch_grating_ft(wavenumber=k_d,
patch_diameter=d)
network.set_stimulus(stimulus)
network.compute_response(ganglion, recompute_ft=True)
response[i, j] = ganglion.center_response[0]
assert (abs(response - response_e) < 1e-10).all()
def test_nonlagged_x_cells():
R_g_e = np.array([36.8000000000, 105.5231755992, 80.5607155005, 60.1117829282, 56.7461598311, 56.5031395446, 56.4951231072, 56.4950008690, 56.4950000029, 56.4950000000])/pq.s
R_r_e = np.array([9.1000000000, 48.9839667670, 20.1915848052, 11.3884937007, 13.2064761119, 13.9269478889, 14.0176607852, 14.0235290468, 14.0237452461, 14.0237499389])/pq.s
patch_diameter = np.linspace(0, 14, 10) * pq.deg
R_g = np.zeros(len(patch_diameter)) / pq.s
R_r = np.zeros(len(patch_diameter)) / pq.s
network = pylgn.Network()
integrator = network.create_integrator(nt=1, nr=7, dt=1*pq.ms, dr=0.2*pq.deg)
ganglion = network.create_ganglion_cell(background_response=36.8/pq.s)
relay = network.create_relay_cell(background_response=9.1/pq.s)
Wg_r = spl.create_dog_ft(A=-1, a=0.62*pq.deg, B=-0.85, b=1.26*pq.deg)
Krig_r = spl.create_gauss_ft(A=1, a=0.88*pq.deg)
Krg_r = spl.create_delta_ft()
ganglion.set_kernel((Wg_r, tpl.create_delta_ft()))
network.connect(ganglion, relay, (Krg_r, tpl.create_delta_ft()), weight=0.81)
network.connect(ganglion, relay, (Krig_r, tpl.create_delta_ft()), weight=-0.56)
for i, d in enumerate(patch_diameter):
stimulus = pylgn.stimulus.create_patch_grating_ft(patch_diameter=d, contrast=-131.3)
network.set_stimulus(stimulus)
network.compute_response(ganglion, recompute_ft=True)
network.compute_response(relay, recompute_ft=True)
R_g[i] = ganglion.center_response[0]
R_r[i] = relay.center_response[0]
assert (abs(R_g - R_g_e) < 1e-9).all()
assert (abs(R_r - R_r_e) < 1e-9).all()
def test_edog_spot_response():
response_e = np.array([[0.0000000000, 0.0000000000],
[0.5255489103, 0.3077428042],
[0.1824324695, 0.0174985410],
[0.1505469275, 0.0632031066],
[0.1500017999, 0.0601939846]]) / pq.s
fb_weights = [0, -1.5]
patch_diameter = np.linspace(0, 6, 5) * pq.deg
response = np.zeros([len(patch_diameter), len(fb_weights)]) / pq.s
for j, w_c in enumerate(fb_weights):
network = pylgn.Network()
integrator = network.create_integrator(nt=1, nr=7, dt=1*pq.ms, dr=0.1*pq.deg)
delta_t = tpl.create_delta_ft()
delta_s = spl.create_delta_ft()
Wg_r = spl.create_dog_ft(A=1, a=0.25*pq.deg, B=0.85, b=0.83*pq.deg)
Krc_r = spl.create_gauss_ft(A=1, a=0.83*pq.deg)
ganglion = network.create_ganglion_cell(kernel=(Wg_r, delta_t))
relay = network.create_relay_cell()
cortical = network.create_cortical_cell()
network.connect(ganglion, relay, (delta_s, delta_t), 1.0)
network.connect(cortical, relay, (Krc_r, delta_t), w_c)
network.connect(relay, cortical, (delta_s, delta_t), 1.0)
for i, d in enumerate(patch_diameter):
stimulus = pylgn.stimulus.create_patch_grating_ft(wavenumber=0./pq.deg,
patch_diameter=d)
network.set_stimulus(stimulus)
network.compute_response(relay, recompute_ft=True)
response[i, j] = relay.center_response[0]
network.clear()
assert (abs(response - response_e) < 1e-10).all()
def test_edog_patch_grating_response():
response_e = np.array([[0.0000000000, 0.0000000000],
[0.5694814680, 0.3933539871],
[0.4933473520, 0.3005140715],
[0.5038172723, 0.2842527074],
[0.5039615741, 0.2843698815]]) / pq.s
fb_weights = [0, -1.5]
patch_diameter = np.linspace(0, 6, 5) * pq.deg
response = np.zeros([len(patch_diameter), len(fb_weights)]) / pq.s
for j, w_c in enumerate(fb_weights):
network = pylgn.Network()
integrator = network.create_integrator(nt=1, nr=7, dt=1*pq.ms, dr=0.1*pq.deg)
delta_t = tpl.create_delta_ft()
delta_s = spl.create_delta_ft()
Wg_r = spl.create_dog_ft(A=1, a=0.25*pq.deg, B=0.85, b=0.83*pq.deg)
Krc_r = spl.create_gauss_ft(A=1, a=0.83*pq.deg)
ganglion = network.create_ganglion_cell(kernel=(Wg_r, delta_t))
relay = network.create_relay_cell()
cortical = network.create_cortical_cell()
network.connect(ganglion, relay, (delta_s, delta_t), 1.0)
network.connect(cortical, relay, (Krc_r, delta_t), w_c)
network.connect(relay, cortical, (delta_s, delta_t), 1.0)
for i, d in enumerate(patch_diameter):
stimulus = pylgn.stimulus.create_patch_grating_ft(wavenumber=integrator.spatial_angular_freqs[4], patch_diameter=d)
network.set_stimulus(stimulus)
network.compute_response(relay, recompute_ft=True)
response[i, j] = relay.center_response[0]
network.clear()
assert (abs(response - response_e) < 1e-10).all()
import pylgn
import pylgn.kernels.spatial as spl
import pylgn.kernels.temporal as tpl
import quantities as pq
# create network
network = pylgn.Network()
# create integrator
integrator = network.create_integrator(nt=8,
nr=9,
dt=1 * pq.ms,
dr=0.1 * pq.deg)
# create kernels
Wg_r = spl.create_dog_ft()
Wg_t = tpl.create_biphasic_ft()
# create neurons
ganglion = network.create_ganglion_cell(kernel=(Wg_r, Wg_t))
# create stimulus
stimulus = pylgn.stimulus.create_natural_image(filenames="natural_scene.png",
delay=40 * pq.ms,
duration=80 * pq.ms)
network.set_stimulus(stimulus, compute_fft=True)
# compute
network.compute_response(ganglion)
# visulize
def create_dynamicnetwork(params=None):
"""
Create PyLGN network with temporal kernels.
Parameters
----------
params : None, dict, str
passed to util._parse_parameters
"""
params = util.parse_parameters(params)
# network
network = pylgn.Network()
integrator = network.create_integrator(nt=params['nt'],
nr=params['nr'],
dt=params['dt'],
dr=params['dr'])
# neurons
ganglion = network.create_ganglion_cell()
relay = network.create_relay_cell()
cortical = network.create_cortical_cell()
# RGC impulse-response function
Wg_r = spl.create_dog_ft(A=params['A_g'],
a=params['a_g'],
B=params['B_g'],
b=params['b_g'])
Wg_t = tpl.create_biphasic_ft(phase=params['phase_g'],
damping=params['damping_g'])
ganglion.set_kernel((Wg_r, Wg_t))
# excitatory FF connection
Krg_r = spl.create_gauss_ft(A=params['A_rg_ex'], a=params['a_rg_ex'])
Krg_t = tpl.create_exp_decay_ft(tau=params['tau_rg_ex'],
delay=params['delay_rg_ex'])
network.connect(ganglion, relay, (Krg_r, Krg_t), weight=params['w_rg_ex'])
# inhibitory FF
Krig_r = spl.create_gauss_ft(A=params['A_rg_in'], a=params['a_rg_in'])
Krig_t = tpl.create_exp_decay_ft(tau=params['tau_rg_in'],
delay=params['delay_rg_in'])
network.connect(ganglion,
relay, (Krig_r, Krig_t),
weight=params['w_rg_in'])
# excitatory FB
Krc_ex_r = spl.create_gauss_ft(A=params['A_rc_ex'], a=params['a_rc_ex'])
Krc_ex_t = tpl.create_exp_decay_ft(tau=params['tau_rc_ex'],
delay=params['delay_rc_ex'])
network.connect(cortical,
relay, (Krc_ex_r, Krc_ex_t),
weight=params['w_rc_ex'])
# inhibitory FB
Krc_in_r = spl.create_gauss_ft(A=params['A_rc_in'], a=params['a_rc_in'])
Krc_in_t = tpl.create_exp_decay_ft(tau=params['tau_rc_in'],
delay=params['delay_rc_in'])
network.connect(cortical,
relay, (Krc_in_r, Krc_in_t),
weight=params['w_rc_in'])
# TC feed-forward
Kcr_r = spl.create_delta_ft()
Kcr_t = tpl.create_delta_ft()
network.connect(relay, cortical, (Kcr_r, Kcr_t), weight=params['w_cr'])
return network
def create_staticnetwork(params=None):
"""
Create a PyLGN network where all temporal kernels are delta functions.
Parameters
----------
params : None, dict, str
passed to util._parse_parameters
"""
params = util.parse_parameters(params)
# network
network = pylgn.Network()
integrator = network.create_integrator(nt=0,
nr=params['nr'],
dt=params['dt'],
dr=params['dr'])
# neurons
ganglion = network.create_ganglion_cell()
relay = network.create_relay_cell()
cortical = network.create_cortical_cell()
# RGC impulse-response function
delta_t = tpl.create_delta_ft()
Wg_r = spl.create_dog_ft(A=params['A_g'],
a=params['a_g'],
B=params['B_g'],
b=params['b_g'])
ganglion.set_kernel((Wg_r, delta_t))
# excitatory FF connection
Krg_r = spl.create_gauss_ft(A=params['A_rg_ex'], a=params['a_rg_ex'])
network.connect(ganglion,
relay, (Krg_r, delta_t),
weight=params['w_rg_ex'])
# inhibitory FF
Krig_r = spl.create_gauss_ft(A=params['A_rg_in'], a=params['a_rg_in'])
network.connect(ganglion,
relay, (Krig_r, delta_t),
weight=params['w_rg_in'])
# excitatory FB
Krc_ex_r = spl.create_gauss_ft(A=params['A_rc_ex'], a=params['a_rc_ex'])
network.connect(cortical,
relay, (Krc_ex_r, delta_t),
weight=params['w_rc_ex'])
# inhibitory FB
Krc_in_r = spl.create_gauss_ft(A=params['A_rc_in'], a=params['a_rc_in'])
network.connect(cortical,
relay, (Krc_in_r, delta_t),
weight=params['w_rc_in'])
# TC feed-forward
Kcr_r = spl.create_delta_ft()
network.connect(relay, cortical, (Kcr_r, delta_t), weight=1)
return network
def test_spl_dog_ft():
dog_ft = spl.create_dog_ft(A=1.0, a=0.25, B=0.85, b=0.83, dx=0, dy=0)
assert abs(dog_ft(kx=0.5, ky=1.1)-0.316423256919) < complex(1e-12, 0)
assert abs(dog_ft(kx=1.5, ky=0.1)-0.389361200098) < complex(1e-12, 0)
# list to store spatiotemporal summation curves for each weight
responses = []
for w_fb in fb_weights:
# create network
network = pylgn.Network()
# create integrator
integrator = network.create_integrator(nt=10,
nr=7,
dt=1 * pq.ms,
dr=0.1 * pq.deg)
# create spatial kernels
Wg_s = spl.create_dog_ft(A=1, a=0.62 * pq.deg, B=0.85, b=1.26 * pq.deg)
Krg_s = spl.create_gauss_ft(A=1, a=0.1 * pq.deg)
Kcr_s = spl.create_delta_ft()
Krcr_cen_s = spl.create_gauss_ft(A=1, a=(0.1) * pq.deg)
Krcr_sur_s = spl.create_gauss_ft(A=2, a=(0.9) * pq.deg)
# create temporal kernels
Wg_t = tpl.create_biphasic_ft(phase_duration=42.5 * pq.ms,
damping_factor=0.38,
delay=0 * pq.ms)
Krg_t = tpl.create_exp_decay_ft(18 * pq.ms, delay=0 * pq.ms)
Kcr_t = tpl.create_delta_ft()
Krcr_cen_t = tpl.create_exp_decay_ft(1 * pq.ms, delay=0 *
pq.ms) # decay: 10 ms, delay: 30 ms
Krcr_sur_t = tpl.create_exp_decay_ft(1 * pq.ms, delay=30 *
pq.ms) # decay: 20 ms, delay: 30 ms
def test_core():
dt = 5 * pq.ms
dr = 0.1 * pq.deg
# create network
network = pylgn.Network()
# create integrator
integrator = network.create_integrator(nt=10, nr=7, dt=dt, dr=dr)
# create neurons
ganglion = network.create_ganglion_cell()
relay = network.create_relay_cell(2. / pq.s)
cortical = network.create_cortical_cell(0 / pq.s)
# connect neurons
Wg_r = spl.create_dog_ft(A=1, a=0.62 * pq.deg, B=0.83, b=1.26 * pq.deg)
Wg_t = tpl.create_biphasic_ft(phase=43 * pq.ms,
damping=0.38,
delay=0 * pq.ms)
Krg_r = spl.create_delta_ft(shift_x=0 * pq.deg, shift_y=0 * pq.deg)
Krg_t = tpl.create_delta_ft(delay=0 * pq.ms)
Krc_r = spl.create_dog_ft(A=1 * 0.9,
a=0.1 * pq.deg,
B=2 * 0.9,
b=0.9 * pq.deg)
Krc_t = tpl.create_delta_ft(delay=20 * pq.ms)
Kcr_r = spl.create_delta_ft(shift_x=0 * pq.deg, shift_y=0 * pq.deg)
Kcr_t = tpl.create_delta_ft(delay=0 * pq.ms)
ganglion.set_kernel((Wg_r, Wg_t))
network.connect(ganglion, relay, (Krg_r, Krg_t))
network.connect(cortical, relay, (Krc_r, Krc_t))
network.connect(relay, cortical, (Kcr_r, Kcr_t))
print(ganglion.annotations["kernel"], "\n")
# create stimulus
k_g = integrator.spatial_angular_freqs[2]
w_g = -integrator.temporal_angular_freqs[40]
# stimulus = pylgn.stimulus.create_patch_grating_ft(angular_freq=w_g,
# wavenumber=k_g,
# orient=0.0,
# patch_diameter=3,
# contrast=4)
stimulus = pylgn.stimulus.create_flashing_spot_ft(patch_diameter=4 *
pq.deg,
contrast=4,
delay=4 * pq.ms,
duration=20 * pq.ms)
network.set_stimulus(stimulus, compute_fft=False)
# network.set_stimulus(stimulus)
# print(pylgn.closure_params(stimulus))
#
# # compute
# network.compute_irf(relay)
network.compute_response(relay)
# network.compute_response(ganglion)
#
# # write
# print("shape cube: ", relay.irf_ft.shape)
# print("Unit irf:", relay.irf.units)
# # print("Units irf_ft: ", relay.irf_ft.units)
# print("Units resp: ", relay.response.units)
# visulize
# import matplotlib.pyplot as plt
# # import matplotlib.animation as animation
# # plt.imshow(relay.irf[0, :, :].real)
#
# plt.figure()
# # plt.plot(integrator.times, relay.response[:, 64, 64])
# w_g = w_g.rescale(pq.Hz) * integrator.times/integrator.times.max()
# plt.plot(w_g/2./np.pi, relay.response[:, 64, 64])
# plt.show()
# pylgn.plot.animate_cube(relay.response)
# clear network
network.clear()
