def run_power_test(actual_mx, disc, red, blue, n, s, annuli, count,
two_level_sample):
r_frac = len(red)
b_frac = len(blue)
power_count = 0
for _ in range(count):
pts = red + blue
random.shuffle(pts)
new_red = pts[:r_frac]
new_blue = pts[b_frac:]
m_sample = pyscan.my_sample(new_red, s)
b_sample = pyscan.my_sample(new_blue, s)
if two_level_sample:
net_set1 = pyscan.my_sample(m_sample, n)
net_set2 = pyscan.my_sample(b_sample, n)
else:
net_set1 = m_sample
net_set2 = b_sample
n = s
net_set = net_set1 + net_set2
m_sample = pyscan.to_weighted(m_sample)
b_sample = pyscan.to_weighted(b_sample)
reg, mx = pyscan.max_annuli_scale(net_set, m_sample, b_sample, annuli,
disc)
if mx > actual_mx:
power_count += 1
return power_count / count
def testing_flux_framework(output_file,
red,
blue,
l_s,
h_s,
count,
region_name="disk",
two_level_sample=True,
ham_sample=False,
max_time=None):
fieldnames = [
"disc", "region", "n", "s", "time", "m_disc", "m_disc_approx"
]
with open(output_file, 'w') as f:
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
for i in np.logspace(l_s, h_s, count):
eps = i
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
disc = utils.disc_to_func("disc")
start_time = time.time()
m_sample = [
pyscan.WPoint(1.0, p[0], p[1], 1.0)
for p in pyscan.my_sample(red, s)
]
b_sample = [
pyscan.WPoint(1.0, p[0], p[1], 1.0)
for p in pyscan.my_sample(blue, s)
]
if two_level_sample:
net_set1 = pyscan.my_sample(m_sample, n)
net_set2 = pyscan.my_sample(b_sample, n)
if ham_sample:
s = int(1 / (2 * eps**(4.0 / 3)) *
math.log(1 / eps)**(2 / 3.0))
m_sample = pyscan.ham_tree_sample(m_sample, s)
b_sample = pyscan.ham_tree_sample(b_sample, s)
else:
net_set1 = [pyscan.Point(p[0], p[1], p[2]) for p in m_sample]
net_set2 = [pyscan.Point(p[0], p[1], p[2]) for p in b_sample]
n = s
net_set1 = [pyscan.Point(p[0], p[1], p[2]) for p in net_set1]
net_set2 = [pyscan.Point(p[0], p[1], p[2]) for p in net_set2]
net_set = net_set1 + net_set2
if region_name == "halfplane":
reg, mx = pyscan.max_halfplane(net_set, m_sample, b_sample,
disc)
elif region_name == "disk":
reg, mx = pyscan.max_disk(net_set, m_sample, b_sample, disc)
elif region_name == "rectangle":
grid = pyscan.Grid(n, m_sample, b_sample)
s1 = pyscan.max_subgrid_linear(grid, -1.0, 1.0)
s2 = pyscan.max_subgrid_linear(grid, 1.0, -1.0)
if s1.fValue() > s2.fValue():
reg = grid.toRectangle(s1)
mx = s1.fValue()
else:
reg = grid.toRectangle(s2)
mx = s2.fValue()
else:
return
end_time = time.time()
st = time.time()
actual_mx = pyscan.evaluate_range(reg, red, blue, disc)
et = time.time()
print("Time to evaluate region {}".format(et - st))
row = {
"disc": "disc",
"region": region_name,
"n": n,
"s": s,
"time": end_time - start_time,
"m_disc_approx": mx,
"m_disc": actual_mx
}
writer.writerow(row)
f.flush()
print(row)
if max_time is not None and end_time - start_time > max_time:
return
def testing_full_framework(
trajectories,
l_s, h_s, count,
vparam="eps",
eps=.01,
r=.04,
p=0.5,
q=.2,
alpha=.01,
planted_points=None,
actual_mx=None,
max_disk_r=.1,
min_disk_r=.05,
disc_name="disc",
input_size=10000):
"""
How do I convert the trajectories over?
1) Just sample evenly from the length.
2) Choose points evenly
3) Choose
:param trajectories:
:param l_s:
:param h_s:
:param count:
:param vparam:
:param eps:
:param eps_r:
:param r:
:param q:
:param disc_name:
:param region_name:
:param input_size:
:return:
"""
output_file = "{0}_multi_disk_{1:.2f}_{2:.2f}_full_discrepancy.csv".format(disc_name, min_disk_r, max_disk_r)
fieldnames = ["vparam", "input_size", "region", "disc", "n", "s", "r", "p", "q", "alpha", "time",
"m_disc",
"m_disc_approx"]
with open(output_file, 'w') as f:
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
for i in np.logspace(l_s, h_s, count):
if vparam == "eps":
eps = i
elif vparam == "r":
r = i
elif vparam == "q":
q = i
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
disc = utils.disc_to_func(disc_name)
st = time.time()
if planted_points is None:
red, blue, _, actual_mx = pyscan.plant_full_disk(trajectories, r, p, q, disc)
else:
red, blue = planted_points
red_sample = pyscan.my_sample(red, s)
blue_sample = pyscan.my_sample(blue, s)
red_net = pyscan.my_sample(red, n)
blue_net = pyscan.my_sample(blue, n)
net = red_net + blue_net
et = time.time()
print("Time to plant region {}".format(et - st))
start_time = time.time()
reg, mx = pyscan.max_disk_traj_grid(net, [pyscan.WTrajectory(1.0, traj) for traj in red_sample],
[pyscan.WTrajectory(1.0, traj) for traj in blue_sample], min_disk_r, max_disk_r, disc)
end_time = time.time()
row = {"vparam": vparam,
"input_size": input_size,
"disc": disc_name,
"region": "multi_disk",
"n": n, "s": s,
"r": r, "q": q,"p":p,
"alpha":alpha,
"time": end_time - start_time,
"m_disc_approx": mx,
"m_disc": actual_mx}
writer.writerow(row)
f.flush()
print(row)
def testing_partial_framework(red,
blue,
output_file,
l_s,
h_s,
count,
r=.04,
p=0.5,
q=.2,
error_thresh=3,
two_level_sample=True,
ham_sample=True,
disc_name="disc",
region_name="disk",
sample_method="block",
max_time=None):
"""
How do I convert the trajectories over?
1) Just sample evenly from the length.
2) Choose points evenly
3) Choose
:param trajectories:
:param l_s:
:param h_s:
:param count:
:param vparam:
:param eps:
:param eps_r:
:param r:
:param q:
:param disc_name:
:param region_name:
:param input_size:
:return:
"""
fieldnames = [
"disc", "region", "n", "s", "r", "p", "q", "time", "m_disc",
"m_disc_approx", "sample_method"
]
with open(output_file, 'w') as f:
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
disc = utils.disc_to_func(disc_name)
s_prime = int(10**(2 * error_thresh) + .5)
m_sample_prime = pyscan.uniform_sample(red, s_prime, False)
b_sample_prime = pyscan.uniform_sample(blue, s_prime, False)
for i in np.logspace(l_s, h_s, count):
eps = i
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
start_time = time.time()
if sample_method == "block":
f_sample = pyscan.block_sample
elif sample_method == "even":
f_sample = pyscan.even_sample
elif sample_method == "uniform":
f_sample = pyscan.uniform_sample
m_sample = f_sample(red, s, False)
b_sample = f_sample(blue, s, False)
m_sample = pyscan.to_weighted(m_sample)
b_sample = pyscan.to_weighted(b_sample)
if two_level_sample:
net_set1 = pyscan.my_sample(m_sample, n)
net_set2 = pyscan.my_sample(b_sample, n)
if ham_sample:
s = int(1 / (2 * eps**(4.0 / 3)) *
math.log(1 / eps)**(2 / 3.0))
m_sample = pyscan.ham_tree_sample(m_sample, s)
b_sample = pyscan.ham_tree_sample(b_sample, s)
else:
net_set1 = [pyscan.Point(p[0], p[1], p[2]) for p in m_sample]
net_set2 = [pyscan.Point(p[0], p[1], p[2]) for p in b_sample]
n = s
net_set1 = [pyscan.Point(p[0], p[1], p[2]) for p in net_set1]
net_set2 = [pyscan.Point(p[0], p[1], p[2]) for p in net_set2]
net_set = net_set1 + net_set2
if region_name == "halfplane":
reg, mx = pyscan.max_halfplane(net_set, m_sample, b_sample,
disc)
elif region_name == "disk":
reg, mx = pyscan.max_disk(net_set, m_sample, b_sample, disc)
elif region_name == "rectangle":
grid = pyscan.Grid(n, m_sample, b_sample)
s1 = pyscan.max_subgrid_linear(grid, -1.0, 1.0)
s2 = pyscan.max_subgrid_linear(grid, 1.0, -1.0)
if s1.fValue() > s2.fValue():
reg = grid.toRectangle(s1)
mx = s1.fValue()
else:
reg = grid.toRectangle(s2)
mx = s2.fValue()
else:
return
end_time = time.time()
actual_mx = pyscan.evaluate_range(
reg, pyscan.to_weighted(m_sample_prime),
pyscan.to_weighted(b_sample_prime), disc)
row = {
"disc": disc_name,
"region": region_name,
"n": n,
"s": s,
"r": r,
"q": q,
"p": p,
"time": end_time - start_time,
"m_disc_approx": mx,
"m_disc": actual_mx,
"sample_method": sample_method
}
writer.writerow(row)
print(row)
f.flush()
if max_time is not None and end_time - start_time > max_time:
return
def testing_bandwidth_framework(pts,
output_file,
l_s,
h_s,
count,
r=.04,
p=0.8,
q=.5,
eps=.02,
annuli_eps=.1,
two_level_sample=True,
algorithm="annuli",
statistic="k",
power_test=None,
max_time=None,
seed=None):
if seed is not None:
random.seed(a=seed)
else:
seed = random.randint(0, 10000000)
random.seed(a=seed)
#seed = 9704595
red, blue, bandwidth_orig, center_pt = pyscan.plant_kernel_disk_region(
pts, r, p, q)
actual_mx = pyscan.disc_bernoulli_kern(red, blue, p, q, bandwidth_orig,
center_pt)
try:
with open(output_file, "r") as fh:
exists = True
except FileNotFoundError:
exists = False
with open(output_file, 'a+') as f:
writer = None
for i in np.logspace(l_s, h_s, count):
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
print(i)
bandwidth = bandwidth_orig * i
start_time = time.time()
m_sample = pyscan.my_sample(red, s)
b_sample = pyscan.my_sample(blue, s)
if two_level_sample:
net_set1 = pyscan.my_sample(m_sample, n)
net_set2 = pyscan.my_sample(b_sample, n)
else:
net_set1 = m_sample
net_set2 = b_sample
n = s
net_set = net_set1 + net_set2
m_sample = pyscan.to_weighted(m_sample)
b_sample = pyscan.to_weighted(b_sample)
# kern = pyscan.gaussian_kernel(bandwidth)
# disc = pyscan.Bernoulli_K(kern)
r_g = bandwidth * math.sqrt(math.e) * eps * 5
disk_r = bandwidth * math.sqrt(math.log(1 / eps))
if algorithm == "kernel":
reg, mx = pyscan.max_kernel_slow(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel2":
reg, mx = pyscan.max_kernel_slow2(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_adaptive":
reg, mx = pyscan.max_kernel_adaptive(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_prune":
disk_r = bandwidth * math.sqrt(
math.log((len(m_sample) + len(b_sample)) / eps))
reg, mx = pyscan.max_kernel_prune_far(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_fast":
reg, mx = pyscan.max_kernel(m_sample, b_sample, r_g, disk_r,
bandwidth)
end_time = time.time()
print(reg, mx)
(p_m, q_m, mx) = pyscan.measure_kernel(reg.get_origin(),
pyscan.to_weighted(red),
pyscan.to_weighted(blue),
bandwidth_orig)
row = {
"disc":
"bernoulli",
"n":
n,
"s":
s,
"r":
r,
"q":
q,
"p":
p,
"p_m":
p_m,
"q_m":
q_m,
"time":
end_time - start_time,
"m_disc_approx":
mx,
"m_disc":
actual_mx,
"center_distance":
reg.get_origin().dist(center_pt),
"kl_div":
Kl_divergence(pts, bandwidth_orig, center_pt,
reg.get_origin()),
"l2_dist":
l2_dist(pts, bandwidth_orig, center_pt, reg.get_origin()),
"angular_dist":
angular_dist(pts, bandwidth_orig, center_pt, reg.get_origin()),
"jc":
extended_jc(pts, bandwidth_orig, center_pt, reg.get_origin()),
"power":
None,
"bandwidth":
bandwidth_orig,
"seed":
seed,
"scale":
i
}
if writer is None:
writer = csv.DictWriter(f, fieldnames=list(row.keys()))
if not exists:
writer.writeheader()
writer.writerow(row)
print(row)
f.flush()
if max_time is not None and end_time - start_time > max_time:
return
def testing_partial_framework(pts,
output_file,
l_s,
h_s,
count,
r=.04,
p=0.8,
q=.5,
annuli_eps=.1,
two_level_sample=True,
algorithm="annuli",
statistic="k",
power_test=None,
max_time=None,
seed=None,
planted_reg=None):
if seed is not None:
random.seed(a=seed)
else:
seed = random.randint(0, 10000000)
random.seed(a=seed)
#seed = 9704595
if planted_reg is None:
red, blue, bandwidth, center_pt = pyscan.plant_kernel_disk_region(
pts, r, p, q)
else:
red, blue, bandwidth, center_pt = planted_reg
bernoulli_sample = [(pt, 1) for pt in red] + [(pt, 0) for pt in blue]
print(bandwidth, center_pt)
actual_mx = pyscan.disc_bernoulli_kern(red, blue, p, q, bandwidth,
center_pt)
try:
with open(output_file, "r") as fh:
exists = True
except FileNotFoundError:
exists = False
with open(output_file, 'a+') as f:
writer = None
for i in np.logspace(l_s, h_s, count):
eps = i
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
start_time = time.time()
if statistic == "bs":
baseline_sample = pyscan.my_sample(bernoulli_sample, s)
m_sample = [pt for (pt, id) in baseline_sample if id]
b_sample = [pt for (pt, _) in baseline_sample]
print(len(m_sample))
print(len(baseline_sample))
print("here")
else:
m_sample = pyscan.my_sample(red, s)
b_sample = pyscan.my_sample(blue, s)
if two_level_sample:
net_set1 = pyscan.my_sample(m_sample, n)
net_set2 = pyscan.my_sample(b_sample, n)
else:
net_set1 = m_sample
net_set2 = b_sample
n = s
net_set = net_set1 + net_set2
m_sample = pyscan.to_weighted(m_sample)
b_sample = pyscan.to_weighted(b_sample)
# kern = pyscan.gaussian_kernel(bandwidth)
# disc = pyscan.Bernoulli_K(kern)
r_g = bandwidth * math.sqrt(math.e) * eps * 5
disk_r = bandwidth * math.sqrt(math.log(1 / eps))
if algorithm == "kernel":
reg, mx = pyscan.max_kernel_slow(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel2":
reg, mx = pyscan.max_kernel_slow2(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_adaptive":
reg, mx = pyscan.max_kernel_adaptive(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_prune":
disk_r = bandwidth * math.sqrt(
math.log((len(m_sample) + len(b_sample)) / eps))
reg, mx = pyscan.max_kernel_prune_far(m_sample, b_sample, r_g,
disk_r, bandwidth)
elif algorithm == "kernel_fast":
reg, mx = pyscan.max_kernel(m_sample, b_sample, r_g, disk_r,
bandwidth)
elif algorithm == "combinatorial" or "satscan" in algorithm:
if statistic == "k":
disc_f = pyscan.RKULLDORF
elif statistic == "b":
disc_f = pyscan.bernoulli(0)
elif statistic == "bs":
disc_f = pyscan.rbernoulli(eps * .0001)
#disc_f = pyscan.bernoulli(len(red), len(blue), eps / 2)
elif statistic == "d":
disc_f = pyscan.DISC
if algorithm == "combinatorial":
reg, mx = pyscan.max_disk(net_set, m_sample, b_sample, disc_f)
elif algorithm == "satscan_grid":
reg, mx = pyscan.satscan_grid(m_sample, b_sample, r_g, disk_r,
disc_f)
elif algorithm == "satscan_points":
reg, mx = pyscan.satscan_points(m_sample, b_sample, disc_f)
end_time = time.time()
print("Finished this run.")
print(reg, mx)
(p_m, q_m, mx) = pyscan.measure_kernel(reg.get_origin(),
pyscan.to_weighted(red),
pyscan.to_weighted(blue),
bandwidth)
if power_test is not None:
power = run_power_test(mx, disc, red, blue, n, s, annuli,
power_test, two_level_sample)
else:
power = None
# r_g = bandwidth * math.sqrt(math.e) * eps * 5
# disk_r = bandwidth * math.sqrt(math.log((len(m_sample) + len(b_sample)) / eps))
# centers = pyscan.kernel_centers(m_sample, b_sample, r_g, disk_r, bandwidth)
#
# fig, ax = plt.subplots(figsize=(20,20))
#
# bx = Mapping.get_bx(pts)
# #ax, _ = Mapping.map_trajectories([], [], bx)
# pyscan.plot_points(ax, m_sample, "r")
# pyscan.plot_points(ax, b_sample, "b")
# #pyscan.plot_points(ax, net_set, "k")
# #pyscan.plot_kernel(ax, pts, center_pt, bandwidth, res=50)
# #pyscan.plot_points(ax, centers, "k")
# print(reg.get_origin())
# pyscan.plot_kernel(ax, pts, reg.get_origin(), bandwidth, res=50)
# plt.axis('off')
# plt.show()
#kl_val = Kl_divergence(pts, bandwidth, center_pt, reg.get_origin())
kl_val = 0
print("Finished kl div")
l2 = l2_dist(pts, bandwidth, center_pt, reg.get_origin())
print("Finished l2")
angle = 0 #angular_dist(pts, bandwidth, center_pt, reg.get_origin())
print("Finished Angle")
ejc = extended_jc(pts, bandwidth, center_pt, reg.get_origin())
print("Finished EJC")
row = {
"disc": "bernoulli",
"n": n,
"s": s,
"r": r,
"q": q,
"p": p,
"p_m": p_m,
"q_m": q_m,
"time": end_time - start_time,
"m_disc_approx": mx,
"m_disc": actual_mx,
"center_distance": reg.get_origin().dist(center_pt),
"kl_div": kl_val,
"l2_dist": l2,
"angular_dist": angle,
"jc": ejc,
"power": power,
"bandwidth": bandwidth,
"seed": seed
}
if writer is None:
writer = csv.DictWriter(f, fieldnames=list(row.keys()))
if not exists:
writer.writeheader()
writer.writerow(row)
print("Run time {}".format(time.time() - start_time))
print(row)
f.flush()
if max_time is not None and end_time - start_time > max_time:
return
return trajectory_pieces
import paths
eps = .1
s = 800
r = .08
p = .5
q = .8
pts = paths.load_philly()
red, blue, bandwidth, center_pt = pyscan.plant_kernel_disk_region(pts, r, p, q)
bx = get_bx(pts)
m_sample = pyscan.my_sample(red, s)
b_sample = pyscan.my_sample(blue, s)
# r_g = bandwidth * math.sqrt(math.e) * eps * 5
# disk_r = bandwidth * math.sqrt(math.log(1 / eps))
#reg, mx = pyscan.max_kernel_slow(pyscan.to_weighted(m_sample), pyscan.to_weighted(b_sample), r_g, disk_r, bandwidth)
ax, _ = map_trajectories([], [], bx)
plot_points(ax, m_sample, "r", transform=projection)
plot_points(ax, b_sample, "b", transform=projection)
pyscan.plot_kernel(ax, pts, center_pt, bandwidth, res=50, transform=projection)
#pyscan.plot_kernel(ax, pts, reg.get_origin(), bandwidth, res=50, transform=projection)
#plt.show()
plt.savefig("philly.pdf", bbox_inches='tight')
def plot_trajectory_set(trajectories, sample):
ax = plt.subplot()
for traj in pyscan.my_sample(trajectories, sample):
plot_line(ax, traj, "r")
plt.show()
def testing_full_framework(
red, blue,
output_file,
l_s, h_s, count,
vparam="eps",
eps=.01,
alpha=.01,
max_disk_r=None,
min_disk_r=None,
disc_name="disc",
region_name="halfplane",
sample_method="halfplane",
fast_disk = True,
two_level_sample=True,
max_time=None):
"""
How do I convert the trajectories over?
1) Just sample evenly from the length.
2) Choose points evenly
3) Choose
:param trajectories:
:param l_s:
:param h_s:
:param count:
:param vparam:
:param eps:
:param eps_r:
:param r:
:param q:
:param disc_name:
:param region_name:
:param input_size:
:return:
"""
fieldnames = ["vparam", "disc", "region", "n", "s", "n_pts", "m_pts", "b_pts", "alpha", "time",
"m_disc", "m_disc_approx", "sample_method"]
with open(output_file, 'w') as f:
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
for i in np.logspace(l_s, h_s, count):
if vparam == "eps":
eps = i
elif vparam == "alpha":
alpha = i
n = 1 / eps
s = 1 / (2 * eps * eps)
n = int(round(n) + .1)
s = int(round(s) + .1)
disc = utils.disc_to_func(disc_name)
red_sample = pyscan.my_sample(red, s)
blue_sample = pyscan.my_sample(blue, s)
if two_level_sample:
red_net = pyscan.my_sample(red, n)
blue_net = pyscan.my_sample(blue, n)
else:
red_net = red_sample
blue_net = blue_sample
net = red_net + blue_net
print("Running: {} {}".format(n, s))
start_time = time.time()
if region_name == "multiscale_disk":
if max_disk_r is not None and alpha > max_disk_r:
print("Max Disk Radius is greater than alpha")
continue
reg, mx = multiscale_disk(min_disk_r, max_disk_r, alpha, red_sample, blue_sample, net, disc, fast_disk)
m_sample, b_sample, net_set = [], [], []
else:
if sample_method == "halfplane":
m_sample = [pyscan.halfplane_kernel([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj], alpha) for traj in red_sample]
b_sample = [pyscan.halfplane_kernel([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj], alpha) for traj in blue_sample]
pt_net = [pyscan.halfplane_kernel([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj], alpha) for traj in net]
elif sample_method == "dp":
m_sample = [pyscan.dp_compress(traj, alpha) for traj in red_sample]
b_sample = [pyscan.dp_compress(traj, alpha) for traj in blue_sample]
pt_net = [pyscan.dp_compress(traj, alpha) for traj in net]
elif sample_method == "hull":
m_sample = [pyscan.convex_hull([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj]) for traj in red_sample]
b_sample = [pyscan.convex_hull([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj]) for traj in blue_sample]
pt_net = [pyscan.convex_hull([pyscan.Point(pt[0], pt[1], 1.0) for pt in traj]) for traj in net]
elif sample_method is None:
#just takes the waypoints.
m_sample = [[pyscan.Point(pt[0], pt[1], 1.0) for pt in traj] for traj in red_sample]
b_sample = [[pyscan.Point(pt[0], pt[1], 1.0) for pt in traj] for traj in blue_sample]
pt_net = [[pyscan.Point(pt[0], pt[1], 1.0) for pt in traj] for traj in net]
elif sample_method == "grid":
m_sample = [pyscan.grid_kernel(traj, alpha) for traj in red_sample]
b_sample = [pyscan.grid_kernel(traj, alpha) for traj in blue_sample]
pt_net = [pyscan.grid_kernel(traj, alpha) for traj in net]
elif sample_method == "lifting":
m_sample = [pyscan.lifting_kernel(traj, alpha) for traj in red_sample]
b_sample = [pyscan.lifting_kernel(traj, alpha) for traj in blue_sample]
pt_net = [pyscan.lifting_kernel(traj, alpha) for traj in net]
elif sample_method == "grid_direc":
if max_disk_r is not None and alpha > max_disk_r:
print("Max Disk Radius is greater than alpha")
continue
chord_l = math.sqrt(4 * alpha * max(min_disk_r, alpha) - 2 * alpha * alpha)
m_sample = [pyscan.grid_direc_kernel(pyscan.dp_compress(traj, alpha), chord_l, alpha) for traj in
red_sample]
b_sample = [pyscan.grid_direc_kernel(pyscan.dp_compress(traj, alpha), chord_l, alpha) for traj in
blue_sample]
pt_net = [pyscan.grid_direc_kernel(pyscan.dp_compress(traj, alpha), chord_l, alpha) for traj in net]
elif sample_method == "even":
m_sample = [pyscan.even_sample_error(traj, alpha, False) for traj in red_sample]
b_sample = [pyscan.even_sample_error(traj, alpha, False) for traj in blue_sample]
pt_net = [pyscan.even_sample_error(traj, alpha, False) for traj in net]
else:
return
if region_name == "multiscale_disk_fixed":
m_sample = list(pyscan.trajectories_to_labels(m_sample))
b_sample = list(pyscan.trajectories_to_labels(b_sample))
net_set = list(pyscan.trajectories_to_labels(pt_net))
reg, mx = multiscale_disk_fixed(min_disk_r, max_disk_r, m_sample, b_sample, net_set, disc, fast_disk)
else:
m_sample = list(pyscan.trajectories_to_labels(m_sample))
b_sample = list(pyscan.trajectories_to_labels(b_sample))
net_set = list(itertools.chain.from_iterable(pt_net))
if region_name == "halfplane":
reg, mx = pyscan.max_halfplane_labeled(net_set, m_sample, b_sample, disc)
elif region_name == "disk":
reg, mx = pyscan.max_disk_labeled(net_set, m_sample, b_sample, disc)
elif region_name == "rectangle":
reg, mx = pyscan.max_rect_labeled(n, 2 * max_disk_r, m_sample, b_sample, disc)
elif region_name == "rectangle_scale":
reg, mx = pyscan.max_rect_labeled_scale(n, 2 * max_disk_r, alpha, net_set, m_sample, b_sample, disc)
else:
return
end_time = time.time()
actual_mx = pyscan.evaluate_range_trajectory(reg, red, blue, disc)
row = {"vparam": vparam,
"disc": disc_name,
"region": region_name,
"n": n, "s": s,
"n_pts": len(net_set), "m_pts":len(m_sample), "b_pts":len(b_sample),
"alpha":alpha,
"time": end_time - start_time,
"m_disc_approx": mx,
"m_disc": actual_mx,
"sample_method": sample_method}
writer.writerow(row)
f.flush()
print(row)
if max_time is not None and end_time - start_time > max_time:
return