def test_power1d_ordinary_freq():
p = [0] * 40
for i in range(40):
pb = PowerBox(8001, dim=1, pk=lambda k: 1.0 * k**-3., boxlength=1.0)
p[i], k = get_power(pb.delta_x(), pb.boxlength)
assert np.allclose(np.mean(np.array(p), axis=0)[2000:],
1.0 * k[2000:]**-3.,
rtol=2)
def test_resolution():
var = [0] * 6
for i in range(6):
pb = PowerBox(64 * 2**i,
dim=2,
pk=lambda k: 1.0 * k**-2.,
boxlength=1.0,
angular_freq=True)
var[i] = np.var(gaussian_filter(pb.delta_x(), sigma=2**i, mode='wrap'))
print(var / var[0])
assert np.allclose(var / var[0], 1, atol=1e-2)
class TestDirect(object):
def setup_method(self, test_method):
self.pb = PowerBox(N, lambda k: k**-2., dim=1)
def test_hermitian(self):
ensure_hermitian(self.pb.delta_k())
def test_reality_elementwise(self):
ensure_reality_elementwise(self.pb.delta_x())
def test_reality(self):
ensure_reality(self.pb.delta_x())
def test_k_zero_ignore():
pb = PowerBox(50, dim=2, pk=lambda k: 1.0 * k**-2., boxlength=1.0, b=1)
dx = pb.delta_x()
p1, k1 = get_power(dx, pb.boxlength, bin_ave=False)
p0, k0 = get_power(dx, pb.boxlength, ignore_zero_mode=True, bin_ave=False)
assert np.all(k1 == k0)
assert np.all(p1[1:] == p0[1:])
assert p1[0] != p0[0]
def test_ln_vs_straight_standard_freq():
# Set up two boxes with exactly the same parameters
pb = PowerBox(128,lambda u : 12.*u**-2., dim=3,seed=1234,boxlength=1200.,a=0,b=2*np.pi)
ln_pb = LogNormalPowerBox(128,lambda u : 12.*u**-2., dim=3,seed=1234,boxlength=1200.,a=0,b=2*np.pi)
pk = get_power(pb.delta_x(),pb.boxlength,a=0,b=2*np.pi)[0]
ln_pk = get_power(ln_pb.delta_x(), pb.boxlength,a=0,b=2*np.pi)[0]
pk = pk[1:-1]
ln_pk = ln_pk[1:-1]
print(np.mean(np.abs((pk-ln_pk)/pk)), np.abs((pk-ln_pk)/pk))
assert np.mean(np.abs((pk-ln_pk)/pk)) < 2e-1 # 10% agreement
def test_power2d():
p = [0] * 5
for i in range(5):
pb = PowerBox(200,
dim=2,
pk=lambda k: 1.0 * k**-2.,
boxlength=1.0,
b=1)
p[i], k = get_power(pb.delta_x(), pb.boxlength, b=1)
assert np.allclose(np.mean(np.array(p), axis=0)[100:],
1.0 * k[100:]**-2.,
rtol=2)
def test_power1d_halfN():
p = [0] * 40
for i in range(40):
pb = PowerBox(4001,
dim=1,
pk=lambda k: 1.0 * k**-3.,
boxlength=1.0,
b=1)
p[i], k = get_power(pb.delta_x(), pb.boxlength, b=1)
assert np.allclose(np.mean(np.array(p), axis=0)[1000:],
1.0 * k[1000:]**-3.,
rtol=2)
def test_k_weights():
pb = PowerBox(50, dim=2, pk=lambda k: 1.0 * k**-2., boxlength=1.0, b=1)
dx = pb.delta_x()
k_weights = np.ones_like(dx)
k_weights[:, 25] = 0
p1, k1 = get_power(dx, pb.boxlength, bin_ave=False)
p0, k0 = get_power(dx, pb.boxlength, bin_ave=False, k_weights=k_weights)
assert np.all(k1 == k0)
assert not np.allclose(p1, p0)
def test_discrete_power_gaussian():
pb = PowerBox(N=512,
dim=2,
boxlength=100.,
pk=lambda u: 0.1 * u**-1.5,
ensure_physical=True)
sample = pb.create_discrete_sample(nbar=1000.)
power, bins = get_power(sample, pb.boxlength, N=pb.N)
res = np.mean(np.abs(power[50:-50] / (0.1 * bins[50:-50]**-1.5) - 1))
print(res)
assert res < 1e-1
def _powerbox(self):
if self.params['use_lognormal']:
return LogNormalPowerBox(N=self.ncells, pk=self.params['power_spectrum'],
dim=2, boxlength=np.max(self.sky_size).value,a=0,b=2*np.pi)
else:
return PowerBox(N=self.ncells, pk=self.params['power_spectrum'],
dim=2, boxlength=np.max(self.sky_size).value,a=0,b=2*np.pi)
class TestDirectEven(TestDirect):
def setup_method(self, test_method):
self.pb = PowerBox(N - 1, lambda k: k**-2., dim=2)
def test_reality_elementwise(self):
return True ## It won't be element-wise correct for even case.
def test_reality(self):
ensure_reality(self.pb.delta_x())
def powerbox(self):
if self.use_lognormal:
return LogNormalPowerBox(N=self.ncells,
pk=self.point_source_power_spec,
dim=2,
boxlength=np.max(self.sky_size.value),
a=self.a,
b=self.b)
else:
return PowerBox(N=self.ncells,
pk=self.point_source_power_spec,
dim=2,
boxlength=np.max(self.sky_size).value,
a=self.a,
b=self.b)
def setup_method(self, test_method):
self.pb = PowerBox(N - 1, lambda k: k**-2., dim=2)
def test_power3d():
pb = PowerBox(50, dim=2, pk=lambda k: 1.0 * k**-2., boxlength=1.0, b=1)
p, k = get_power(pb.delta_x(), pb.boxlength, b=1)
print(p / (1.0 * k**-2.))
assert np.allclose(p, 1.0 * k**-2., rtol=2)
