def test_mc_sampling_errors(method, ninvar, seed):
with pytest.raises(ValueError):
sampling(ndraws=512,
method=method,
ninvar=ninvar,
ninvar_max=None,
seed=seed)
def test_mc_sampling(method, ninvar, seed, ans):
pcts = sampling(ndraws=512,
method=method,
ninvar=ninvar,
ninvar_max=None,
seed=seed)
assert pcts[0] == pytest.approx(ans)
def draw(self,
ninvar_max : int = None,
) -> None:
"""
Perform the random draws based on the sampling method and the
statistical distribution, and map those draws to values.
Parameters
----------
ninvar_max : int
The total number of input variables for the simulation.
"""
self.pcts = []
self.nums = []
dist = self.dist(**self.distkwargs)
if self.firstcaseismedian:
self.ncases = self.ndraws + 1
self.pcts.append(0.5)
self.nums.append(np.array(self.getDistMedian()))
pcts = sampling(ndraws=self.ndraws, method=self.samplemethod,
ninvar=self.ninvar, ninvar_max=ninvar_max,
seed=self.seed)
self.pcts.extend(pcts)
self.nums.extend([np.array(x) for x in dist.ppf(pcts)])
if any(np.isinf(self.nums)):
warn( 'Infinite value drawn. Check distribution and parameters: ' +
f'{self.dist=}, {self.distkwargs=}')
if self.samplemethod in (SampleMethod.SOBOL, SampleMethod.HALTON):
warn(f'Infinite value draw may happen with {self.dist=} for the ' +
f'first point of the {self.samplemethod} sampling method. ' +
f'Consider using {SampleMethod.SOBOL_RANDOM} instead.')
if any(np.isnan(num) for num in self.nums):
raise ValueError( 'Invalid draw. Check distribution and parameters: ' +
f'{self.dist=}, {self.distkwargs=}')
self.mapNums()
1, n + 1))**dimension / np.arange(1, n + 1)
error1sig = error1sig_sobol
# Leading zeros will throw off plots, fill with reasonable dummy data
error1sig[error1sig == 0] = max(error1sig)
error = error1sig * pct2sig(conf)
return error
samplepointsrandom = []
samplepointssobol = []
for i in range(d):
samplepointsrandom.append(
sampling(ndraws=n,
method=SampleMethod.RANDOM,
ninvar=i + 1,
seed=seed + i)) # Need different seed for random draws
samplepointssobol.append(
sampling(ndraws=n,
method=SampleMethod.SOBOL,
ninvar=i + 1,
seed=None,
ninvar_max=d))
# Discrepancy calculations take too long
'''
samplepointsrandom = np.array(samplepointsrandom).transpose()
samplepointsrandom = np.reshape(samplepointsrandom, (n, d))
errrandom = scipy.stats.qmc.discrepancy(samplepointsrandom)
samplepointssobol = np.array(samplepointssobol).transpose()
samplepointssobol = np.reshape(samplepointssobol, (n, d))
def plot_sampling_test(ndraws, method, seeds, genplot=True):
import scipy.stats
pcts = np.array([
sampling(ndraws=ndraws,
method=method,
ninvar=1,
ninvar_max=2,
seed=seeds[0]),
sampling(ndraws=ndraws,
method=method,
ninvar=2,
ninvar_max=2,
seed=seeds[1])
])
if genplot:
fig, axs = plt.subplots(1, 3)
fig.suptitle(f"samplemethod = '{method}'", fontweight='bold')
fig.set_dpi(96)
fig.set_size_inches(16, 5)
axs[0].set_title('Uniform')
axs[0].scatter(pcts[0], pcts[1], alpha=0.5)
square = plt.Rectangle((0, 0), 1, 1, color='k', fill=False)
axs[0].add_patch(square)
axs[0].set_xlim([-0.1, 1.1])
axs[0].set_ylim([-0.1, 1.1])
axs[0].set_aspect('equal')
axs[1].set_title('Normal')
axs[1].scatter(scipy.stats.norm.ppf(pcts[0]),
scipy.stats.norm.ppf(pcts[1]),
alpha=0.5)
circle1 = plt.Circle((0, 0), 1, color='k', fill=False)
circle2 = plt.Circle((0, 0), 2, color='k', fill=False)
circle3 = plt.Circle((0, 0), 3, color='k', fill=False)
axs[1].add_patch(circle1)
axs[1].add_patch(circle2)
axs[1].add_patch(circle3)
axs[1].set_xlim([-3.5, 3.5])
axs[1].set_ylim([-3.5, 3.5])
axs[1].set_aspect('equal')
axs[2].set_title('Frequency Spectra')
ndraws_freq = 10000
# ndraws_freq = 1000000 # For better frequency plots
n_freq_grid = 2**8
f_max = n_freq_grid / 2
pcts_freq = np.array([
sampling(ndraws=ndraws_freq,
method=method,
ninvar=1,
ninvar_max=2,
seed=seeds[0]),
sampling(ndraws=ndraws_freq,
method=method,
ninvar=2,
ninvar_max=2,
seed=seeds[1])
])
pcts_freq_int = np.round(pcts_freq * (n_freq_grid - 1)).astype(int)
S = np.zeros([n_freq_grid, n_freq_grid])
for i in range(ndraws_freq):
S[pcts_freq_int[0, i], pcts_freq_int[1, i]] += 1
FS = np.fft.fft2(S)
axs[2].imshow(np.log(np.abs(np.fft.fftshift(FS))**2),
cmap='Blues_r',
extent=[-f_max, f_max, -f_max, f_max],
aspect="equal")
# fig.savefig(f'../docs/{method}_sampling.png')
return pcts