def plot_gate_ring(sigma=np.deg2rad(13),
shift=np.deg2rad(40),
theta_t=0.,
phi_t=0.,
subplot=111):
theta, phi = fibonacci_sphere(samples=1000, fov=161)
d_cl1 = (np.sin(shift - theta) * np.cos(theta_t) +
np.cos(shift - theta) * np.sin(theta_t) * np.cos(phi - phi_t))
gate = np.power(np.exp(-np.square(d_cl1) / (2. * np.square(sigma))), 1)
ax = plt.subplot(subplot, polar=True)
ax.scatter(phi,
theta,
s=10,
marker='o',
c=gate,
cmap="Reds",
vmin=0,
vmax=1)
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(90)])
plt.yticks([np.deg2rad(28)], [r'$28^\circ$'])
ax.set_xticks(np.linspace(0, 2 * np.pi, 8, endpoint=False))
ax.set_xticklabels([
r'$0^\circ$', r'$45^\circ$', r'$90^\circ$', r'$135^\circ$',
r'$180^\circ$', r'$-135^\circ$', r'$-90^\circ$', r'$-45^\circ$'
])
return plt
def plot_res2ele(samples=1000, noise=0., subplot=111):
ele, azi, azi_diff, res = [], [], [], []
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
phi_s = phi_s[theta_s > np.pi / 18]
theta_s = theta_s[theta_s > np.pi / 18]
samples = theta_s.size
for e, a in zip(theta_s, phi_s):
d_err, d_eff, tau, _, _ = evaluate(sun_azi=a, sun_ele=e, tilting=False, noise=noise)
azi.append(a)
ele.append(e)
res.append(tau.flatten())
ele = np.rad2deg(ele).flatten()
res = (np.array(res).flatten() - 1.06) * 7 / 4
ele = ele[res <= 2]
res = res[res <= 2]
ele_pred = 26 * (1 - 2 * np.arcsin(1 - res) / np.pi) + 15 # + np.random.randn(res.size)
plt.subplot(subplot)
plt.scatter(res, ele, c='black', marker='.')
plt.scatter(res, ele_pred, c='red', marker='.')
plt.plot([-.5, 3 * np.pi / 4], [18.75, 18.75], "k--")
plt.plot([-.5, 3 * np.pi / 4], [65.98, 65.98], "k--")
plt.ylabel(r'$\epsilon (\circ)$')
plt.xlabel(r'$\tau$')
plt.xlim([-.5, 3 * np.pi / 4])
plt.ylim([90, 0])
# plt.xticks([0, 90, 180, 270, 360])
return plt
def plot_gate_cost(samples=500, **kwargs):
d_err, d_eff, tau, _, _ = evaluate(samples=samples, tilting=True, **kwargs)
tau = np.rad2deg(tau)
d_mean = np.nanmean(d_err)
d_se = d_err.std() / np.sqrt(d_err.size)
print "Tilt overall 0 deg 30 deg 60 deg "
print "---------------------------------------------------------------------------------------"
print "Mean cost %.2f +/- %.4f" % (d_mean, d_se),
if samples == 1000:
samples /= 2
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
d_00 = d_err[:, 0]
d_30 = np.nanmean(d_err[:, 1:9], axis=1)
d_60 = np.nanmean(d_err[:, 9:], axis=1)
print " %.2f +/- %.4f" % (np.nanmean(d_00), np.nanstd(d_00) / d_00.size),
print " %.2f +/- %.4f" % (np.nanmean(
d_err[:, 1:9]), np.nanstd(d_err[:, 1:9]) / d_err[:, 1:9].size),
print " %.2f +/- %.4f" % (np.nanmean(
d_err[:, 9:]), np.nanstd(d_err[:, 9:]) / d_err[:, 9:].size)
for i, ang, dd in zip(range(3), [0, np.pi / 6, np.pi / 3],
[d_00, d_30, d_60]):
ax = plt.subplot2grid((1, 10), (0, i * 3), colspan=3, polar=True)
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
plt.scatter(phi_s,
np.rad2deg(theta_s),
marker=".",
c=dd,
cmap="Reds",
vmin=0,
vmax=90)
plt.scatter(np.pi,
np.rad2deg(ang),
marker="o",
c="yellowgreen",
edgecolors="black")
plt.text(-np.deg2rad(50), 145, ["A", "B", "C"][i], fontsize=12)
plt.axis("off")
plt.subplot2grid((3, 10), (1, 9))
plt.imshow(np.array([np.arange(0, np.pi / 2, np.pi / 180)] * 10).T,
cmap="Reds")
plt.xticks([])
plt.yticks([0, 45, 89], [r"0", r"$\frac{\pi}{4}$", r"$\geq\frac{\pi}{2}$"])
return plt
def plot_res2ele(samples=1000, noise=0., subplot=111):
ele, azi, azi_diff, res = [], [], [], []
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
phi_s = phi_s[theta_s > np.pi / 18]
theta_s = theta_s[theta_s > np.pi / 18]
samples = theta_s.size
for e, a in zip(theta_s, phi_s):
d_err, d_eff, tau, _, _ = evaluate(sun_azi=a,
sun_ele=e,
tilting=False,
noise=noise)
azi.append(a)
ele.append(e)
res.append(tau.flatten())
ele = np.rad2deg(ele).flatten()
res = np.array(res).flatten()
ele_pred = 26 * (
1 - 2 * np.arcsin(np.clip(2.855 - 3.5 * res, -1, 1)) / np.pi) + 15
plt.subplot(subplot)
plt.scatter(res, ele, c='black', marker='.')
plt.scatter(np.clip(res, 0, 4), ele_pred, c='red', marker='.')
plt.plot([-.5, 3 * np.pi / 4], [18.75, 18.75], "k--")
plt.plot([-.5, 3 * np.pi / 4], [65.98, 65.98], "k--")
plt.ylabel(r'$\epsilon (\circ)$')
plt.xlabel(r'$\tau$')
plt.xticks([0.5, 0.75, 1, 1.25])
plt.xlim([.43, 1.21])
plt.ylim([90, 0])
return plt
def __init__(self, nb_lenses=60, fov=60, nb_tb1=8, consider_tilting=False,
b_thetas=True, b_phis=True, b_alphas=True, b_ws=True, noise=0.5):
# Initialise the position and orientation of the lenses with fibonacci distribution
thetas, phis = fibonacci_sphere(samples=nb_lenses, fov=fov)
thetas = (thetas + np.pi) % (2 * np.pi) - np.pi
phis = (phis + np.pi) % (2 * np.pi) - np.pi
alphas = (phis + 3 * np.pi / 2) % (2 * np.pi) - np.pi
# initialise weights of the computational model
phi_tb1 = np.linspace(0., 2 * np.pi, nb_tb1, endpoint=False) # TB1 preference angles
w = nb_tb1 / (2. * nb_lenses) * np.sin(phi_tb1[np.newaxis] - alphas[:, np.newaxis])
# create initial feature-vector
self.x_init = self.vectorise(thetas, phis, alphas, w)
self.theta_init = thetas
self.phi_init = phis
self.w_f = lambda a: nb_tb1 / (2. * nb_lenses) * np.maximum(np.sin(phi_tb1[np.newaxis] - a[:, np.newaxis]), 0.)
self.ndim = (3 + nb_tb1) * nb_lenses # number of features
self.lb = np.hstack(( # lower bound of parameters
np.zeros(nb_lenses), # theta
np.full(nb_lenses, -np.pi), # phi
np.full(nb_lenses, -np.pi), # alpha
-np.ones(nb_tb1 * nb_lenses) # weights
))
self.ub = np.hstack(( # upper bound of parameters
np.full(nb_lenses, np.deg2rad(fov) / 2), # theta
np.full(nb_lenses, np.pi), # phi
np.full(nb_lenses, np.pi), # alpha
np.ones(nb_tb1 * nb_lenses) # weights
))
self._thetas = b_thetas
self._phis = b_phis
self._alphas = b_alphas
self._ws = b_ws
self.__consider_tilting = consider_tilting
self._noise = noise
vmin=-np.pi,
vmax=np.pi)
ax.scatter(phi_s, theta_s, s=100, edgecolor='black', facecolor='yellow')
# ax.scatter(sky.phi_t + np.pi, sky.theta_t, s=200, edgecolor='black', facecolor='greenyellow')
ax.set_ylim([0, np.pi / 2])
ax.set_yticks([])
ax.set_xticks(np.linspace(0, 2 * np.pi, 8, endpoint=False))
ax.set_xticklabels([
r'$0^\circ$ (N)', r'$45^\circ$ (NE)', r'$90^\circ$ (E)',
r'$135^\circ$ (SE)', r'$180^\circ$ (S)', r'$-135^\circ$ (SW)',
r'$-90^\circ$ (W)', r'$-45^\circ$ (NW)'
])
plt.show()
if __name__ == "__main__":
from compoundeye.geometry import fibonacci_sphere
t, p = fibonacci_sphere(10000, 180)
s = Sky(theta_s=np.pi / 6)
# s = sky_from_observer()
# s = Sky.from_type(11)
# s.theta_s = np.pi/6
y, p, a = s(t, p)
visualise_luminance(s)
# visualise_degree_of_polarisation(s)
# visualise_angle_of_polarisation(s)
n = 60
omega = 56
theta, phi, fit = angles_distribution(n, float(omega))
theta_t, phi_t = 0., 0.
# sun position
seville = ephem.Observer()
seville.lat = '37.392509'
seville.lon = '-5.983877'
seville.date = datetime(2018, 6, 21, 9, 0, 0)
sun = ephem.Sun()
sun.compute(seville)
theta_s = np.array([np.pi / 2 - sun.alt])
phi_s = np.array([(sun.az + np.pi) % (2 * np.pi) - np.pi])
theta_sky, phi_sky = fibonacci_sphere(1000, 180)
# ant-world
noise_type = "canopy"
mode = "uneven"
noise = 0.0
ttau = .06
dx = 1e-02
# world = load_world()
routes = load_routes()
flow = dx * np.ones(2) / np.sqrt(2)
max_theta = 0.
for ni, noise in enumerate([0.0, 0.2, 0.4, 0.6, 0.8, 1.0]):
# for ni, noise in enumerate([0.0]):
def skyfeatures(noise=0., simple_pol=False, samples=1000, verbose=False):
# default parameters
tau_L = 2.
c1 = .6
c2 = 4.
eps = np.finfo(float).eps
AA, BB, CC, DD, EE = SensorObjective.T_L.dot(np.array(
[tau_L, 1.])) # sky parameters
T_T = np.linalg.pinv(SensorObjective.T_L)
tau_L, c = T_T.dot(np.array([AA, BB, CC, DD, EE]))
tau_L /= c # turbidity correction
# Prez. et. al. Luminance function
def L(cchi, zz):
ii = zz < (np.pi / 2)
ff = np.zeros_like(zz)
if zz.ndim > 0:
ff[ii] = (1. + AA * np.exp(BB / (np.cos(zz[ii]) + eps)))
elif ii:
ff = (1. + AA * np.exp(BB / (np.cos(zz) + eps)))
pphi = (1. + CC * np.exp(DD * cchi) + EE * np.square(np.cos(cchi)))
return ff * pphi
theta_s, phi_s = np.array([np.pi / 6]), np.array([np.pi])
theta, phi = fibonacci_sphere(samples, 180)
phi = phi[theta <= np.pi / 2]
theta = theta[theta <= np.pi / 2]
samples = theta.size
theta = (theta - np.pi) % (2 * np.pi) - np.pi
phi = (phi + np.pi) % (2 * np.pi) - np.pi
alpha = (phi + np.pi / 2) % (2 * np.pi) - np.pi
# SKY INTEGRATION
gamma = np.arccos(
np.cos(theta) * np.cos(theta_s) +
np.sin(theta) * np.sin(theta_s) * np.cos(phi - phi_s))
# Intensity
I_prez, I_00, I_90 = L(gamma, theta), L(0., theta_s), L(
np.pi / 2, np.absolute(theta_s - np.pi / 2))
# influence of sky intensity
I = (1. / (I_prez + eps) - 1. /
(I_00 + eps)) * I_00 * I_90 / (I_00 - I_90 + eps)
chi = (4. / 9. - tau_L / 120.) * (np.pi - 2 * theta_s)
Y_z = (4.0453 * tau_L - 4.9710) * np.tan(chi) - 0.2155 * tau_L + 2.4192
Y = np.maximum(Y_z * I_prez / (I_00 + eps), 0.) # Illumination
# Degree of Polarisation
M_p = np.exp(-(tau_L - c1) / (c2 + eps))
LP = np.square(np.sin(gamma)) / (1 + np.square(np.cos(gamma)))
if simple_pol:
P = np.clip(2. / np.pi * M_p * LP, 0., 1.)
else:
P = np.clip(
2. / np.pi * M_p * LP * (theta * np.cos(theta) +
(np.pi / 2 - theta) * I), 0., 1.)
# Angle of polarisation
_, A = tilt(theta_s, phi_s + np.pi, theta, phi)
# create cloud disturbance
if type(noise) is np.ndarray:
if noise.size == P.size:
# print "yeah!"
eta = np.array(noise, dtype=bool)
else:
eta = np.zeros_like(theta, dtype=bool)
eta[:noise.size] = noise
elif noise > 0:
eta = np.argsort(np.absolute(
np.random.randn(*P.shape)))[:int(noise * P.shape[0])]
# eta = np.array(np.absolute(np.random.randn(*P.shape)) < noise, dtype=bool)
if verbose:
print "Noise level: %.4f (%.2f %%)" % (noise, 100. * eta.sum() /
float(eta.size))
else:
eta = np.zeros_like(theta, dtype=bool)
P[eta] = 0. # destroy the polarisation pattern
Y[eta] = 10.
return Y, P, A, theta, phi
eta = np.zeros_like(theta, dtype=bool)
P[eta] = 0. # destroy the polarisation pattern
Y[eta] = 10.
return Y, P, A, theta, phi
if __name__ == "__main__":
from compoundeye import POLCompassDRA
from environment import Sky
samples = 1000
noise = .0
# noise = .99
theta, phi = fibonacci_sphere(samples, 180)
sky = Sky(np.pi / 6, np.pi)
y, p, a = sky(theta, phi)
# y, p, a, theta, phi = skyfeatures(noise, samples=50000)
# dra = POLCompassDRA()
# r = dra(sky, noise=noise)
#
# print r.shape
# print r
#
# plt.figure("pol-%02d" % (10 * noise), figsize=(3, 3))
# ax = plt.subplot(111, polar=True)
# ax.scatter(dra.phi, dra.theta, s=100, marker='.', c=r, cmap="coolwarm", vmin=-.1, vmax=.1)
# ax.set_theta_zero_location("N")
# ax.set_theta_direction(-1)
# ax.set_ylim([0, np.deg2rad(30)])
def azirot(vec, phi):
"""
Rotate a vector horizontally and clockwise.
:param vec: the 3D vector
:param phi: the azimuth of the rotation
"""
Rz = np.asarray([[np.cos(phi), -np.sin(phi), 0],
[np.sin(phi), np.cos(phi), 0], [0, 0, 1]])
return Rz.dot(vec)
if __name__ == "__main__":
from compoundeye.geometry import fibonacci_sphere
# v = np.array([[1], [0], [0]], dtype=float)
# v = np.array([0, 1, 0], dtype=float)
theta_s, phi_s = fibonacci_sphere(samples=1000, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
samples = theta_s.size
theta_t, phi_t = np.pi / 4, np.pi / 4
v_s = sph2vec(theta_s, phi_s, zenith=True)
R = sph2rotmat(theta_t, phi_t)
theta_s_1, phi_s_1, _ = vec2sph(R.dot(v_s), zenith=True)
theta_s_2, phi_s_2 = tilt(theta_t, phi_t, theta_s, phi_s)
print "Elevation:", np.all(np.isclose(theta_s_1, theta_s_2)),
print "--- Azimuth:", np.all(np.isclose(phi_s_1, phi_s_2))
def evaluate_old(
n=60,
omega=56,
noise=0.,
nb_cl1=16,
sigma=np.deg2rad(13),
shift=np.deg2rad(40),
nb_tb1=8,
use_default=False,
weighted=True,
fibonacci=False,
simple_pol=False,
uniform_poliriser=False,
# single evaluation
sun_azi=None,
sun_ele=None,
# data parameters
tilting=True,
samples=1000,
show_plots=False,
show_structure=False,
verbose=False):
# default parameters
tau_L = 2.
c1 = .6
c2 = 4.
eps = np.finfo(float).eps
AA, BB, CC, DD, EE = T_L.dot(np.array([tau_L, 1.])) # sky parameters
T_T = np.linalg.pinv(T_L)
tau_L, c = T_T.dot(np.array([AA, BB, CC, DD, EE]))
tau_L /= c # turbidity correction
# Prez. et. al. Luminance function
def L(cchi, zz):
ii = zz < (np.pi / 2)
ff = np.zeros_like(zz)
if zz.ndim > 0:
ff[ii] = (1. + AA * np.exp(BB / (np.cos(zz[ii]) + eps)))
elif ii:
ff = (1. + AA * np.exp(BB / (np.cos(zz) + eps)))
pphi = (1. + CC * np.exp(DD * cchi) + EE * np.square(np.cos(cchi)))
return ff * pphi
if tilting:
angles = np.array([[0., 0.], [np.pi / 6, 0.], [np.pi / 6, np.pi / 4],
[np.pi / 6, 2 * np.pi / 4],
[np.pi / 6, 3 * np.pi / 4],
[np.pi / 6, 4 * np.pi / 4],
[np.pi / 6, 5 * np.pi / 4],
[np.pi / 6, 6 * np.pi / 4],
[np.pi / 6, 7 * np.pi / 4], [np.pi / 3, 0.],
[np.pi / 3, np.pi / 4], [np.pi / 3, 2 * np.pi / 4],
[np.pi / 3, 3 * np.pi / 4],
[np.pi / 3, 4 * np.pi / 4],
[np.pi / 3, 5 * np.pi / 4],
[np.pi / 3, 6 * np.pi / 4],
[np.pi / 3, 7 * np.pi / 4]]) # 17
if samples == 1000:
samples /= 2
else:
angles = np.array([[0., 0.]]) # 1
# generate the different sun positions
if sun_azi is not None or sun_ele is not None:
theta_s = sun_ele if type(sun_ele) is np.ndarray else np.array(
[sun_ele])
phi_s = sun_azi if type(sun_azi) is np.ndarray else np.array([sun_azi])
else:
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
samples = theta_s.size
# generate the properties of the sensor
try:
theta, phi, fit = angles_distribution(n, float(omega))
except ValueError:
theta = np.empty(0, dtype=np.float32)
phi = np.empty(0, dtype=np.float32)
fit = False
if not fit or n > 100 or fibonacci:
theta, phi = fibonacci_sphere(n, float(omega))
# theta, phi, fit = angles_distribution(n, omega)
# if not fit:
# print theta.shape, phi.shape
theta = (theta - np.pi) % (2 * np.pi) - np.pi
phi = (phi + np.pi) % (2 * np.pi) - np.pi
alpha = (phi + np.pi / 2) % (2 * np.pi) - np.pi
# computational model parameters
phi_cl1 = np.linspace(0., 4 * np.pi, nb_cl1,
endpoint=False) # CL1 preference angles
phi_tb1 = np.linspace(0., 2 * np.pi, nb_tb1,
endpoint=False) # TB1 preference angles
# initialise lists for the statistical data
d = np.zeros((samples, angles.shape[0]), dtype=np.float32)
t = np.zeros_like(d)
d_eff = np.zeros((samples, angles.shape[0]), dtype=np.float32)
a_ret = np.zeros_like(t)
tb1 = np.zeros((samples, angles.shape[0], nb_tb1), dtype=np.float32)
# iterate through the different tilting angles
for j, (theta_t, phi_t) in enumerate(angles):
# transform relative coordinates
theta_s_, phi_s_ = tilt(theta_t, phi_t, theta=theta_s, phi=phi_s)
theta_, phi_ = tilt(theta_t, phi_t + np.pi, theta=theta, phi=phi)
_, alpha_ = tilt(theta_t, phi_t + np.pi, theta=np.pi / 2, phi=alpha)
for i, (e, a, e_org,
a_org) in enumerate(zip(theta_s_, phi_s_, theta_s, phi_s)):
# SKY INTEGRATION
gamma = np.arccos(
np.cos(theta_) * np.cos(e_org) +
np.sin(theta_) * np.sin(e_org) * np.cos(phi_ - a_org))
# Intensity
I_prez, I_00, I_90 = L(gamma, theta_), L(0., e_org), L(
np.pi / 2, np.absolute(e_org - np.pi / 2))
# influence of sky intensity
I = (1. / (I_prez + eps) - 1. /
(I_00 + eps)) * I_00 * I_90 / (I_00 - I_90 + eps)
chi = (4. / 9. - tau_L / 120.) * (np.pi - 2 * e_org)
Y_z = (4.0453 * tau_L -
4.9710) * np.tan(chi) - 0.2155 * tau_L + 2.4192
if uniform_poliriser:
Y = np.maximum(np.full_like(I_prez, Y_z), 0.)
else:
Y = np.maximum(Y_z * I_prez / (I_00 + eps), 0.) # Illumination
# Degree of Polarisation
M_p = np.exp(-(tau_L - c1) / (c2 + eps))
LP = np.square(np.sin(gamma)) / (1 + np.square(np.cos(gamma)))
if uniform_poliriser:
P = np.ones_like(LP)
elif simple_pol:
P = np.clip(2. / np.pi * M_p * LP, 0., 1.)
else:
P = np.clip(
2. / np.pi * M_p * LP * (theta_ * np.cos(theta_) +
(np.pi / 2 - theta_) * I), 0., 1.)
# Angle of polarisation
if uniform_poliriser:
A = np.full_like(P, a_org + np.pi)
else:
_, A = tilt(e_org, a_org + np.pi, theta_, phi_)
# create cloud disturbance
if noise > 0:
eta = np.absolute(np.random.randn(*P.shape)) < noise
if verbose:
print "Noise level: %.4f (%.2f %%)" % (
noise, 100. * eta.sum() / float(eta.size))
P[eta] = 0. # destroy the polarisation pattern
else:
eta = np.zeros(1)
# COMPUTATIONAL MODEL
# Input (POL) layer -- Photo-receptors
s_1 = 15. * (np.square(np.sin(A - alpha_)) +
np.square(np.cos(A - alpha_)) * np.square(1. - P))
s_2 = 15. * (np.square(np.cos(A - alpha_)) +
np.square(np.sin(A - alpha_)) * np.square(1. - P))
r_1, r_2 = np.sqrt(s_1), np.sqrt(s_2)
# r_1, r_2 = np.log(s_1 + 1.), np.log(s_2 + 1.)
r_pol = (r_1 - r_2) / (r_1 + r_2 + eps)
# Tilting (CL1) layer
d_cl1 = (
np.sin(shift - theta) * np.cos(theta_t) +
np.cos(shift - theta) * np.sin(theta_t) * np.cos(phi - phi_t))
gate = np.power(
np.exp(-np.square(d_cl1) / (2. * np.square(sigma))), 1)
w_cl1 = float(nb_cl1) / float(n) * np.sin(
alpha[:, np.newaxis] - phi_cl1[np.newaxis]) * gate[:,
np.newaxis]
r_cl1 = r_pol.dot(w_cl1)
# Output (TB1) layer
w_tb1 = float(nb_tb1) / float(
2 * nb_cl1) * np.cos(phi_tb1[np.newaxis] -
phi_cl1[:, np.newaxis])
r_tb1 = r_cl1.dot(w_tb1)
if use_default:
w = -float(nb_tb1) / (2. * float(n)) * np.sin(phi_tb1[
np.newaxis] - alpha[:, np.newaxis]) * gate[:, np.newaxis]
r_tb1 = r_pol.dot(w)
# decode response - FFT
R = r_tb1.dot(
np.exp(-np.arange(nb_tb1) * (0. + 1.j) * np.pi /
(float(nb_tb1) / 2.)))
a_pred = (np.pi - np.arctan2(R.imag, R.real)) % (
2. * np.pi) - np.pi # sun azimuth (prediction)
tau_pred = np.absolute(R) # certainty of prediction
d[i, j] = np.absolute(
azidist(np.array([e, a]), np.array([0., a_pred])))
t[i, j] = tau_pred if weighted else 1.
a_ret[i, j] = a_pred
tb1[i, j] = r_tb1
# effective degree of polarisation
M = r_cl1.max() - r_cl1.min()
# M = t[i, j] * 2.
p = np.power(10, M / 2.)
d_eff[i, j] = np.mean((p - 1.) / (p + 1.))
if show_plots:
plt.figure("sensor-noise-%2d" %
(100. * eta.sum() / float(eta.size)),
figsize=(18, 4.5))
ax = plt.subplot(1, 12, 10)
plt.imshow(w_cl1, cmap="coolwarm", vmin=-1, vmax=1)
plt.xlabel("CBL", fontsize=16)
plt.xticks([0, 15], ["1", "16"])
plt.yticks([0, 59], ["1", "60"])
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 6, 6) # , sharey=ax)
plt.imshow(w_tb1, cmap="coolwarm", vmin=-1, vmax=1)
plt.xlabel("TB1", fontsize=16)
plt.xticks([0, 7], ["1", "8"])
plt.yticks([0, 15], ["1", "16"])
cbar = plt.colorbar(ticks=[-1, 0, 1])
cbar.ax.set_yticklabels([r'$\leq$ -1', r'0', r'$\geq$ 1'])
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 1, polar=True)
ax.scatter(phi,
theta,
s=150,
c=r_pol,
marker='o',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(a,
e,
s=100,
marker='o',
edgecolor='black',
facecolor='yellow')
ax.scatter(phi_t + np.pi,
theta_t,
s=200,
marker='o',
edgecolor='black',
facecolor='yellowgreen')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(40)])
ax.set_yticks([])
ax.set_xticks(
np.linspace(-3 * np.pi / 4,
5 * np.pi / 4,
8,
endpoint=False))
ax.set_title("POL Response", fontsize=16)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 2, polar=True)
ax.scatter(phi,
theta,
s=150,
c=r_pol * gate,
marker='o',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(a,
e,
s=100,
marker='o',
edgecolor='black',
facecolor='yellow')
ax.scatter(phi_t + np.pi,
theta_t,
s=200,
marker='o',
edgecolor='black',
facecolor='yellowgreen')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(40)])
ax.set_yticks([])
ax.set_xticks(
np.linspace(-3 * np.pi / 4,
5 * np.pi / 4,
8,
endpoint=False))
ax.set_title("Gated Response", fontsize=16)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 3, polar=True)
x = np.linspace(0, 2 * np.pi, 721)
# CBL
ax.fill_between(x,
np.full_like(x, np.deg2rad(60)),
np.full_like(x, np.deg2rad(90)),
facecolor="C1",
alpha=.5,
label="CBL")
ax.scatter(phi_cl1[:nb_cl1 / 2] - np.pi / 24,
np.full(nb_cl1 / 2, np.deg2rad(75)),
s=600,
c=r_cl1[:nb_cl1 / 2],
marker='o',
edgecolor='red',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(phi_cl1[nb_cl1 / 2:] + np.pi / 24,
np.full(nb_cl1 / 2, np.deg2rad(75)),
s=600,
c=r_cl1[nb_cl1 / 2:],
marker='o',
edgecolor='green',
cmap="coolwarm",
vmin=-1,
vmax=1)
for ii, pp in enumerate(phi_cl1[:nb_cl1 / 2] - np.pi / 24):
ax.text(pp - np.pi / 20,
np.deg2rad(75),
"%d" % (ii + 1),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
for ii, pp in enumerate(phi_cl1[nb_cl1 / 2:] + np.pi / 24):
ax.text(pp + np.pi / 20,
np.deg2rad(75),
"%d" % (ii + 9),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
# TB1
ax.fill_between(x,
np.full_like(x, np.deg2rad(30)),
np.full_like(x, np.deg2rad(60)),
facecolor="C2",
alpha=.5,
label="TB1")
ax.scatter(phi_tb1,
np.full_like(phi_tb1, np.deg2rad(45)),
s=600,
c=r_tb1,
marker='o',
edgecolor='blue',
cmap="coolwarm",
vmin=-1,
vmax=1)
for ii, pp in enumerate(phi_tb1):
ax.text(pp,
np.deg2rad(35),
"%d" % (ii + 1),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
ax.arrow(pp,
np.deg2rad(35),
0,
np.deg2rad(10),
fc='k',
ec='k',
head_width=.1,
overhang=.3)
# Sun position
ax.scatter(a,
e,
s=500,
marker='o',
edgecolor='black',
facecolor='yellow')
# Decoded TB1
# ax.plot([0, a_pred], [0, e_pred], 'k--', lw=1)
ax.plot([0, a_pred], [0, np.pi / 2], 'k--', lw=1)
ax.arrow(a_pred,
0,
0,
np.deg2rad(20),
fc='k',
ec='k',
head_width=.3,
head_length=.2,
overhang=.3)
ax.legend(ncol=2, loc=(-.55, -.1), fontsize=16)
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.pi / 2])
ax.set_yticks([])
ax.set_xticks([])
ax.set_title("Sensor Response", fontsize=16)
plt.subplots_adjust(left=.02, bottom=.12, right=.98, top=.88)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
plt.show()
d_deg = np.rad2deg(d)
if show_structure:
plt.figure("sensor-structure", figsize=(4.5, 4.5))
ax = plt.subplot(111, polar=True)
ax.scatter(phi, theta, s=150, c="black", marker='o')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(50)])
ax.set_yticks([])
ax.set_xticks(
np.linspace(-3 * np.pi / 4, 5 * np.pi / 4, 8, endpoint=False))
ax.set_title("POL Response")
plt.show()
return d_deg, d_eff, t, a_ret, tb1
def evaluate_slow(
n=60,
omega=56,
noise=0.,
nb_cl1=8,
sigma_pol=np.deg2rad(13),
shift_pol=np.deg2rad(40),
nb_tb1=8,
sigma_sol=np.deg2rad(13),
shift_sol=np.deg2rad(40),
use_default=False,
weighted=True,
fibonacci=False,
uniform_polariser=False,
# single evaluation
sun_azi=None,
sun_ele=None,
# data parameters
tilting=True,
samples=1000,
show_plots=False,
show_structure=False,
verbose=False):
if tilting:
angles = np.array([[0., 0.], [np.pi / 6, 0.], [np.pi / 6, np.pi / 4],
[np.pi / 6, 2 * np.pi / 4],
[np.pi / 6, 3 * np.pi / 4],
[np.pi / 6, 4 * np.pi / 4],
[np.pi / 6, 5 * np.pi / 4],
[np.pi / 6, 6 * np.pi / 4],
[np.pi / 6, 7 * np.pi / 4], [np.pi / 3, 0.],
[np.pi / 3, np.pi / 4], [np.pi / 3, 2 * np.pi / 4],
[np.pi / 3, 3 * np.pi / 4],
[np.pi / 3, 4 * np.pi / 4],
[np.pi / 3, 5 * np.pi / 4],
[np.pi / 3, 6 * np.pi / 4],
[np.pi / 3, 7 * np.pi / 4]]) # 17
if samples == 1000:
samples /= 2
else:
angles = np.array([[0., 0.]]) # 1
# generate the different sun positions
if sun_azi is not None or sun_ele is not None:
theta_s = sun_ele if type(sun_ele) is np.ndarray else np.array(
[sun_ele])
phi_s = sun_azi if type(sun_azi) is np.ndarray else np.array([sun_azi])
else:
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi / 2]
theta_s = theta_s[theta_s <= np.pi / 2]
samples = theta_s.size
# generate the properties of the sensor
try:
theta, phi, fit = angles_distribution(n, float(omega))
except ValueError:
theta = np.empty(0, dtype=np.float32)
phi = np.empty(0, dtype=np.float32)
fit = False
if not fit or n > 100 or fibonacci:
theta, phi = fibonacci_sphere(n, float(omega))
# theta, phi, fit = angles_distribution(n, omega)
# if not fit:
# print theta.shape, phi.shape
theta = (theta - np.pi) % (2 * np.pi) - np.pi
phi = (phi + np.pi) % (2 * np.pi) - np.pi
alpha = (phi + np.pi / 2) % (2 * np.pi) - np.pi
# computational model parameters
phi_cl1 = np.linspace(0., 2 * np.pi, nb_cl1,
endpoint=False) # CL1 preference angles
phi_tb1 = np.linspace(0., 2 * np.pi, nb_tb1,
endpoint=False) # TB1 preference angles
# initialise lists for the statistical data
d = np.zeros((samples, angles.shape[0]), dtype=np.float32)
t = np.zeros_like(d)
d_eff = np.zeros((samples, angles.shape[0]), dtype=np.float32)
a_ret = np.zeros_like(t)
tb1 = np.zeros((samples, angles.shape[0], nb_tb1), dtype=np.float32)
# iterate through the different tilting angles
for j, (theta_t, phi_t) in enumerate(angles):
# transform relative coordinates
theta_s_, phi_s_ = tilt(theta_t, phi_t, theta=theta_s, phi=phi_s)
_, alpha_ = tilt(theta_t, phi_t + np.pi, theta=np.pi / 2, phi=alpha)
for i, (e, a, e_org,
a_org) in enumerate(zip(theta_s_, phi_s_, theta_s, phi_s)):
sky = Sky(theta_s=e_org, phi_s=a_org, theta_t=theta_t, phi_t=phi_t)
sky.verbose = verbose
# COMPUTATIONAL MODEL
# Input (POL) layer -- Photo-receptors
dra = POLCompassDRA(n=n, omega=omega)
dra.theta_t = theta_t
dra.phi_t = phi_t
r_pol = dra(sky, noise=noise, uniform_polariser=uniform_polariser)
r_sol = dra.r_po
# Tilting (SOL) layer
d_pol = (np.sin(shift_pol - theta) * np.cos(theta_t) +
np.cos(shift_pol - theta) * np.sin(theta_t) *
np.cos(phi - phi_t))
gate_pol = np.power(
np.exp(-np.square(d_pol) / (2. * np.square(sigma_pol))), 1)
z_pol = -float(nb_cl1) / float(n)
w_cl1_pol = z_pol * np.sin(phi_cl1[
np.newaxis] - alpha[:, np.newaxis]) * gate_pol[:, np.newaxis]
d_sol = (np.sin(shift_sol - theta) * np.cos(theta_t) +
np.cos(shift_sol - theta) * np.sin(theta_t) *
np.cos(phi - phi_t))
gate_sol = np.power(
np.exp(-np.square(d_sol) / (2. * np.square(sigma_sol))), 1)
z_sol = float(nb_cl1) / float(n)
w_cl1_sol = z_sol * np.sin(phi_cl1[
np.newaxis] - alpha[:, np.newaxis]) * gate_sol[:, np.newaxis]
o = 1. / 64.
f_pol, f_sol = .5 * np.power(2 * theta_t / np.pi, o), .5 * (
1 - np.power(2 * theta_t / np.pi, o))
r_cl1_pol = r_pol.dot(w_cl1_pol)
r_cl1_sol = r_sol.dot(w_cl1_sol)
r_cl1 = f_pol * r_cl1_pol + f_sol * r_cl1_sol
# r_cl1 = r_cl1_sol
# Output (TCL) layer
# w_tb1 = np.eye(nb_tb1)
w_tb1 = float(nb_tb1) / float(nb_cl1) * np.cos(
phi_tb1[np.newaxis] - phi_cl1[:, np.newaxis])
r_tb1 = r_cl1.dot(w_tb1)
if use_default:
w = -float(nb_tb1) / (2. * float(n)) * np.sin(
phi_tb1[np.newaxis] -
alpha[:, np.newaxis]) * gate_pol[:, np.newaxis]
r_tb1 = r_pol.dot(w)
# decode response - FFT
R = r_tb1.dot(
np.exp(-np.arange(nb_tb1) * (0. + 1.j) * np.pi /
(float(nb_tb1) / 2.)))
a_pred = (np.pi - np.arctan2(R.imag, R.real)) % (
2. * np.pi) - np.pi # sun azimuth (prediction)
tau_pred = np.maximum(np.absolute(R), 0) # certainty of prediction
d[i, j] = np.absolute(
azidist(np.array([e, a]), np.array([0., a_pred])))
t[i, j] = tau_pred if weighted else 1.
a_ret[i, j] = a_pred
tb1[i, j] = r_tb1
# effective degree of polarisation
M = r_cl1.max() - r_cl1.min()
# M = t[i, j] * 2.
p = np.power(10, M / 2.)
d_eff[i, j] = np.mean((p - 1.) / (p + 1.))
if show_plots:
plt.figure("sensor-noise-%2d" %
(100. * sky.eta.sum() / float(sky.eta.size)),
figsize=(18, 4.5))
ax = plt.subplot(1, 12, 10)
plt.imshow(w_cl1_pol, cmap="coolwarm", vmin=-1, vmax=1)
plt.xlabel("CBL", fontsize=16)
plt.xticks([0, 15], ["1", "16"])
plt.yticks([0, 59], ["1", "60"])
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 6, 6) # , sharey=ax)
plt.imshow(w_tb1, cmap="coolwarm", vmin=-1, vmax=1)
plt.xlabel("TB1", fontsize=16)
plt.xticks([0, 7], ["1", "8"])
plt.yticks([0, 15], ["1", "16"])
cbar = plt.colorbar(ticks=[-1, 0, 1])
cbar.ax.set_yticklabels([r'$\leq$ -1', r'0', r'$\geq$ 1'])
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 1, polar=True)
ax.scatter(phi,
theta,
s=150,
c=r_pol,
marker='o',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(a,
e,
s=100,
marker='o',
edgecolor='black',
facecolor='yellow')
ax.scatter(phi_t + np.pi,
theta_t,
s=200,
marker='o',
edgecolor='black',
facecolor='yellowgreen')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(40)])
ax.set_yticks([])
ax.set_xticks(np.linspace(0, 2 * np.pi, 8, endpoint=False))
ax.set_xticklabels([
r'$0^\circ$', r'$45^\circ$', r'$90^\circ$', r'$135^\circ$',
r'$180^\circ$', r'$-135^\circ$', r'$-90^\circ$',
r'$-45^\circ$'
])
ax.set_title("POL Response", fontsize=16)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 2, polar=True)
ax.scatter(phi,
theta,
s=150,
c=r_pol * gate_pol,
marker='o',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(a,
e,
s=100,
marker='o',
edgecolor='black',
facecolor='yellow')
ax.scatter(phi_t + np.pi,
theta_t,
s=200,
marker='o',
edgecolor='black',
facecolor='yellowgreen')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(40)])
ax.set_yticks([])
ax.set_xticks(np.linspace(0, 2 * np.pi, 8, endpoint=False))
ax.set_xticklabels([
r'$0^\circ$', r'$45^\circ$', r'$90^\circ$', r'$135^\circ$',
r'$180^\circ$', r'$-135^\circ$', r'$-90^\circ$',
r'$-45^\circ$'
])
ax.set_title("Gated Response", fontsize=16)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
ax = plt.subplot(1, 4, 3, polar=True)
x = np.linspace(0, 2 * np.pi, 721)
# CBL
ax.fill_between(x,
np.full_like(x, np.deg2rad(60)),
np.full_like(x, np.deg2rad(90)),
facecolor="C1",
alpha=.5,
label="CBL")
ax.scatter(phi_cl1[:nb_cl1 / 2] - np.pi / 24,
np.full(nb_cl1 / 2, np.deg2rad(75)),
s=600,
c=r_cl1[:nb_cl1 / 2],
marker='o',
edgecolor='red',
cmap="coolwarm",
vmin=-1,
vmax=1)
ax.scatter(phi_cl1[nb_cl1 / 2:] + np.pi / 24,
np.full(nb_cl1 / 2, np.deg2rad(75)),
s=600,
c=r_cl1[nb_cl1 / 2:],
marker='o',
edgecolor='green',
cmap="coolwarm",
vmin=-1,
vmax=1)
for ii, pp in enumerate(phi_cl1[:nb_cl1 / 2] - np.pi / 24):
ax.text(pp - np.pi / 20,
np.deg2rad(75),
"%d" % (ii + 1),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
for ii, pp in enumerate(phi_cl1[nb_cl1 / 2:] + np.pi / 24):
ax.text(pp + np.pi / 20,
np.deg2rad(75),
"%d" % (ii + 9),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
# TB1
ax.fill_between(x,
np.full_like(x, np.deg2rad(30)),
np.full_like(x, np.deg2rad(60)),
facecolor="C2",
alpha=.5,
label="TB1")
ax.scatter(phi_tb1,
np.full_like(phi_tb1, np.deg2rad(45)),
s=600,
c=r_tb1,
marker='o',
edgecolor='blue',
cmap="coolwarm",
vmin=-1,
vmax=1)
for ii, pp in enumerate(phi_tb1):
ax.text(pp,
np.deg2rad(35),
"%d" % (ii + 1),
ha="center",
va="center",
size=10,
bbox=dict(boxstyle="circle", fc="w", ec="k"))
ax.arrow(pp,
np.deg2rad(35),
0,
np.deg2rad(10),
fc='k',
ec='k',
head_width=.1,
overhang=.3)
# Sun position
ax.scatter(a,
e,
s=500,
marker='o',
edgecolor='black',
facecolor='yellow')
# Decoded TB1
# ax.plot([0, a_pred], [0, e_pred], 'k--', lw=1)
ax.plot([0, a_pred], [0, np.pi / 2], 'k--', lw=1)
ax.arrow(a_pred,
0,
0,
np.deg2rad(20),
fc='k',
ec='k',
head_width=.3,
head_length=.2,
overhang=.3)
ax.legend(ncol=2, loc=(-.55, -.1), fontsize=16)
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.pi / 2])
ax.set_yticks([])
ax.set_xticks([])
ax.set_title("Sensor Response", fontsize=16)
plt.subplots_adjust(left=.02, bottom=.12, right=.98, top=.88)
ax.tick_params(axis='both', which='major', labelsize=16)
ax.tick_params(axis='both', which='minor', labelsize=16)
plt.show()
d_deg = np.rad2deg(d)
if show_structure:
plt.figure("sensor-structure", figsize=(4.5, 4.5))
ax = plt.subplot(111, polar=True)
ax.scatter(phi, theta, s=150, c="black", marker='o')
ax.set_theta_zero_location("N")
ax.set_theta_direction(-1)
ax.set_ylim([0, np.deg2rad(50)])
ax.set_yticks([])
ax.set_xticks(
np.linspace(-3 * np.pi / 4, 5 * np.pi / 4, 8, endpoint=False))
ax.set_title("POL Response")
plt.show()
return d_deg, d_eff, t, a_ret, tb1
def _fitness(theta, phi, alpha, w=None, samples=1000, tilt=False,
error=azidist, noise=0., gate_shift=np.deg2rad(55.5), gate_order=7.5,
activation="linear", return_mean=True):
import matplotlib.pyplot as plt
if tilt:
angles = np.array([
[0., 0.],
[np.pi/6, 0.], [np.pi/6, np.pi/4], [np.pi/6, 2*np.pi/4], [np.pi/6, 3*np.pi/4],
[np.pi/6, 4*np.pi/4], [np.pi/6, 5*np.pi/4], [np.pi/6, 6*np.pi/4], [np.pi/6, 7*np.pi/4],
[np.pi/3, 0.], [np.pi/3, np.pi/4], [np.pi/3, 2*np.pi/4], [np.pi/3, 3*np.pi/4],
[np.pi/3, 4*np.pi/4], [np.pi/3, 5*np.pi/4], [np.pi/3, 6*np.pi/4], [np.pi/3, 7*np.pi/4]
]) # 17
if samples == 1000:
samples /= 2
else:
angles = np.array([[0., 0.]]) # 1
theta_s, phi_s = fibonacci_sphere(samples=samples, fov=161)
phi_s = phi_s[theta_s <= np.pi/2]
theta_s = theta_s[theta_s <= np.pi/2]
samples = theta_s.size
d = np.zeros((samples, angles.shape[0]))
d_eff = np.zeros((samples, angles.shape[0]))
nb_tl2, nb_tb1 = 16, 8
phi_tl2 = np.linspace(0., 4 * np.pi, nb_tl2, endpoint=False) # TL2 preference angles
phi_tb1 = np.linspace(0., 2 * np.pi, nb_tb1, endpoint=False) # TB1 preference angles
for j, (theta_t, phi_t) in enumerate(angles):
theta_s_, phi_s_ = transtilt(theta_t, phi_t, theta=theta_s, phi=phi_s)
theta_, phi_ = transtilt(theta_t, phi_t+np.pi, theta=theta, phi=phi)
_, alpha_ = transtilt(theta_t, phi_t+np.pi, theta=np.pi/2, phi=alpha)
if w is None or True:
# Gate + Shift
g = np.power(np.sqrt(1 - np.square(
np.cos(theta + gate_shift) * np.cos(theta_t) +
np.sin(theta + gate_shift) * np.sin(theta_t) * np.cos(phi_t - phi)
# )), gate_order)
)), 0)
# alternative form (w/o shift)
# g = np.power(np.sin(theta_ + shift), order) # dynamic gating
# w_tl2 = 1. / 60. * np.cos(d_tl2)
# w_tl2 = -1. / 60. * np.cos(phi_tl2[np.newaxis] - phi[:, np.newaxis]) * g[..., np.newaxis]
# w_tb1 = float(nb_tb1) / float(nb_tl2) * np.sin(phi_tb1[np.newaxis] - phi_tl2[:, np.newaxis])
w = 8. / (2. * 60.) * np.sin(phi_tb1[np.newaxis] - phi[:, np.newaxis] - np.pi/2) * g[:, np.newaxis]
for i, (e, a, e_org, a_org) in enumerate(zip(theta_s_, phi_s_, theta_s, phi_s)):
_, dop, aop = SensorObjective.encode(e_org, a_org, theta_, phi_)
# ele, azi = SensorObjective.decode(dop, aop, alpha_, w=w, noise=noise)
r = SensorObjective.opticalencode(dop, aop, alpha, noise=noise)
ele, azi = SensorObjective.decode(r, theta, phi, activation=activation,
# w_tl2=w_tl2)
w=w)
d[i, j] = np.absolute(error(np.array([e, a]), np.array([ele, azi])))
# r = np.maximum(r*g, 0)
r = r*g
M = r.max() - r.min()
P = np.power(10, M/2.)
d_eff[i, j] = np.mean((P - 1.) / (P + 1.))
d_deg = np.rad2deg(d)
# plt.figure("Tilts", figsize=(10, 15))
# for i, ang, dd in zip(range(angles.shape[0]), angles, d.T):
# ax = plt.subplot2grid((5, 4), ((i-1) // 4, (i-1) % 4), polar=True)
# ax.set_theta_zero_location("N")
# ax.set_theta_direction(-1)
# plt.scatter(phi_s, np.rad2deg(theta_s), marker=".", c=dd, cmap="Reds", vmin=0, vmax=np.pi/2)
# plt.scatter(ang[1]+np.pi, np.rad2deg(ang[0]), marker="o", c="yellow", edgecolors="black")
# plt.title(r"$\epsilon_t=%03d, \alpha_t=%03d$" % tuple(np.rad2deg(ang)))
# plt.axis("off")
# plt.subplot2grid((5, 4), (4, 0), colspan=3)
# plt.imshow([np.arange(0, np.pi/2, np.pi/180)] * 3, cmap="Reds")
# plt.yticks([])
# plt.xticks([0, 45, 89], [r"0", r"$\frac{\pi}{4}$", r"$\geq\frac{\pi}{2}$"])
# plt.show()
# plt.figure("Tilts", figsize=(10, 3))
# for i, ang, dd in zip(range(3), angles[[0, 1, 9]], d.T[[0, 1, 9]]):
# ax = plt.subplot2grid((1, 10), (0, i * 3), colspan=3, polar=True)
# ax.set_theta_zero_location("N")
# ax.set_theta_direction(-1)
# plt.scatter(phi_s, np.rad2deg(theta_s), marker="o", c=dd, cmap="Reds", vmin=0, vmax=np.pi/2)
# plt.scatter(ang[1]+np.pi, np.rad2deg(ang[0]), marker="o", c="yellowgreen", edgecolors="black")
# plt.text(-np.deg2rad(50), 145, ["A.", "B.", "C."][i], fontsize=12)
# plt.axis("off")
# plt.subplot2grid((3, 10), (1, 9))
# plt.imshow(np.array([np.arange(0, np.pi/2, np.pi/180)] * 10).T, cmap="Reds")
# plt.xticks([])
# plt.yticks([0, 45, 89], [r"0", r"$\frac{\pi}{4}$", r"$\geq\frac{\pi}{2}$"])
# plt.show()
# plt.figure("cost-function")
# wd = np.bartlett(10)
# wd /= wd.sum()
# d_000 = np.convolve(wd, np.rad2deg(d[:, 0]), mode="same")
# plt.plot(np.rad2deg(theta_s), d_000, label=r"$0^\circ$")
# if angles.shape[0] > 1:
# d_030 = np.convolve(wd, np.rad2deg(d[:, 1]), mode="same")
# d_060 = np.convolve(wd, np.rad2deg(d[:, 9]), mode="same")
# plt.plot(np.rad2deg(theta_s), d_030, label=r"$30^\circ$")
# plt.plot(np.rad2deg(theta_s), d_060, label=r"$60^\circ$")
# plt.legend()
# plt.xlim([0, 90])
# plt.ylim([0, 90])
# plt.xlabel(r"sun elevation ($^\circ$)")
# plt.ylabel(r"cost ($^\circ$)")
# plt.show()
if return_mean:
return d_deg.mean()
else:
return d_deg, d_eff, theta_s
