def macroMacroPDF(name, step, pData, varsDir, weight, mcFiles):
import seaborn as sns
from resample import unWeighted_resample
params = {
'lines.linewidth': 1.5,
'backend': 'ps',
'axes.labelsize': 20,
'font.size': 20,
'legend.fontsize': 18,
'xtick.labelsize': 18,
'ytick.labelsize': 18,
'text.usetex': True,
'font.family': 'serif',
'legend.edgecolor': 'k',
'legend.fancybox': False
}
matplotlib.rcParams['text.latex.preamble'] = [r"\usepackage{amsmath}"]
matplotlib.rcParams.update(params)
# read in the original sampled parameter set
labels = [r'$p$' + r' $(\mathrm{MPa})$', r'$q/p$', r'$n$', r'$C^*$']
# Monte Carlo sequence
nSample, numOfObs = weight.shape
enQOIs = np.zeros([4, nSample])
for i in range(nSample):
C, CN, K0, e_r11, e_r21, e_a1, n1, overlap, p1, q1 = np.genfromtxt(
mcFiles[i]).transpose()
enQOIs[0, i] = p1[step]
enQOIs[1, i] = q1[step] / p1[step]
enQOIs[2, i] = n1[step]
enQOIs[3, i] = CN[step]
# importance resampling
ResampleIndices = unWeighted_resample(weight[:, step], nSample * 10)
sortedIndex = np.argsort(weight[:, step])
particles = np.copy(pData)
particles[0, :] /= 1e9
particles[2, :] /= 1e3
particles = particles[:, ResampleIndices]
enQOIs = enQOIs[:, ResampleIndices]
# plot figures
figNO = 0
fig = plt.figure(figsize=(18 / 2.54, 18 / 2.54))
sns.set(style="ticks", rc=matplotlib.rcParams)
for i in range(4):
for j in range(4):
figNO += 1
# plot histograms at the initial and final steps
ax = fig.add_subplot(4, 4, figNO)
if i == j:
# estimate pdf using gaussian kernals
p0, w0 = getPDF(enQOIs[i, :], min(enQOIs[i, :]),
max(enQOIs[i, :]))
ax2 = ax.twinx()
ax2.fill_between(p0, w0, alpha=0.63, facecolor='black')
ax2.tick_params(axis='y', right='off', labelright='off')
ax.set_ylim(min(p0), max(p0))
elif j < i:
cmap = sns.dark_palette("black", as_cmap=True)
ax = sns.kdeplot(enQOIs[j],
enQOIs[i],
cmap=cmap,
kernel='gau',
n_levels=20,
bw='silverman',
linewidths=0.8)
ax.grid()
ax.locator_params(axis='both', tight=True, nbins=2)
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='off',
labelleft='off')
ax.set_xlim(min(enQOIs[j, :]), max(enQOIs[j, :]))
ax.set_ylim(min(enQOIs[i, :]), max(enQOIs[i, :]))
elif j > i:
cmap = sns.dark_palette("black", as_cmap=True)
ax.scatter(enQOIs[j, sortedIndex],
enQOIs[i, sortedIndex],
s=25,
c=weight[sortedIndex, step],
cmap='Greys')
ax.grid()
ax.locator_params(axis='both', tight=True, nbins=2)
ax.set_xlim(min(enQOIs[j, :]), max(enQOIs[j, :]))
ax.set_ylim(min(enQOIs[i, :]), max(enQOIs[i, :]))
if i == 3:
ax.set_xlabel(labels[j], size=params['font.size'], labelpad=10)
ax.tick_params(axis='both',
which='both',
bottom='on',
top='off',
labelbottom='on',
right='off',
left='off',
labelleft='off',
labelsize=params['xtick.labelsize'])
if j == 0:
ax.set_ylabel(labels[i], size=params['font.size'], labelpad=10)
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='on',
labelleft='on',
labelsize=params['xtick.labelsize'])
if i == 3 and j == 0:
ax.tick_params(axis='both',
which='both',
bottom='on',
top='off',
labelbottom='on',
right='off',
left='on',
labelleft='on',
labelsize=params['xtick.labelsize'])
if i != 3 and j != 0:
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='off',
labelleft='off')
plt.tight_layout()
fig.subplots_adjust(left=0.165,
bottom=0.12,
right=0.98,
top=0.99,
hspace=0.1,
wspace=0.1)
plt.savefig('macroMacroUQ_' + varsDir[-8:-1] + '_%i.pdf' % step, dpi=600)
plt.show()
def microMacroPDF(name,
step,
pData,
varsDir,
weight,
mcFiles,
loadWeights=False):
import seaborn as sns
from resample import unWeighted_resample
params = {
'lines.linewidth': 1.5,
'backend': 'ps',
'axes.labelsize': 18,
'font.size': 18,
'legend.fontsize': 16,
'xtick.labelsize': 16,
'ytick.labelsize': 16,
'text.usetex': True,
'font.family': 'serif',
'legend.edgecolor': 'k',
'legend.fancybox': False
}
matplotlib.rcParams['text.latex.preamble'] = [r"\usepackage{amsmath}"]
matplotlib.rcParams.update(params)
# read in the original sampled parameter set
Xlabels = [
r'$E_c$' + r' $(\mathrm{GPa})$', r'$\mu$',
r'$k_m$' + r' $(\mu\mathrm{Nm)}$', r'$\eta_m$'
]
Ylabels = [r'$p$' + r' $(\mathrm{MPa})$', r'$q/p$', r'$n$', r'$C^*$']
if loadWeights:
wPerPair = np.load(varsDir + '/wPerPair%i.npy' % step).item()
else:
wPerPair = {}
# Monte Carlo sequence
nSample, numOfObs = weight.shape
enQOIs = np.zeros([4, nSample])
for i in range(nSample):
C, CN, K0, e_r11, e_r21, e_a1, n1, overlap, p1, q1 = np.genfromtxt(
mcFiles[i]).transpose()
enQOIs[0, i] = p1[step]
enQOIs[1, i] = q1[step] / p1[step]
enQOIs[2, i] = n1[step]
enQOIs[3, i] = CN[step]
# importance resampling
ResampleIndices = unWeighted_resample(weight[:, step], nSample * 10)
particles = np.copy(pData)
particles[0, :] /= 1e9
particles[2, :] /= 1e3
particles = particles[:, ResampleIndices]
enQOIs = enQOIs[:, ResampleIndices]
# remove porosity as QOI if iterNO > 0
if 'iterPF0' not in mcFiles[0]:
enQOIs = np.vstack([enQOIs[:2, :], enQOIs[-1, :]])
for i in [9, 10, 11, 12]:
wPerPair[i] = wPerPair[i + 4]
Ylabels = [r'$p$' + r' $(\mathrm{MPa})$', r'$q/p$', r'$C^*$']
# plot figures
figNO = 0
fig = plt.figure('microMacroUQ_iterPF' + varsDir[-1] + '_%i' % step,
figsize=(12 / 2.54, 12 / 2.54))
sns.set(style="ticks", rc=matplotlib.rcParams)
for i in range(enQOIs.shape[0]):
for j in range(particles.shape[0]):
figNO += 1
# plot histograms at the initial and final steps
ax = fig.add_subplot(enQOIs.shape[0], 4, figNO)
cmap = sns.dark_palette("black", as_cmap=True)
#~ ax = sns.kdeplot(,,cmap=cmap,kernel='gau',cut=3,n_levels=20,bw='silverman',linewidths=0.8); ax.grid()
if loadWeights:
minScore, maxScore, p0, w0 = wPerPair[figNO]
else:
data = np.array([particles[j], enQOIs[i]])
pMin = [min(particles[j]), min(enQOIs[i])]
pMax = [max(particles[j]), max(enQOIs[i])]
minScore, maxScore, p0, w0 = getLogWeightFromGMM(
data.T, pMin, pMax, 1e-2)
wPerPair[figNO] = (minScore, maxScore, p0, w0)
X, Y = p0
plt.contour(X,
Y,
w0,
cmap=cmap,
levels=np.linspace(minScore - abs(minScore) * 0.1,
maxScore, 10),
linewidths=0.7)
plt.grid()
ax.locator_params(axis='both', nbins=2)
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='off',
labelleft='off')
if i == enQOIs.shape[0] - 1:
ax.set_xlabel(Xlabels[j],
size=params['font.size'],
labelpad=10)
ax.tick_params(axis='both',
which='both',
bottom='on',
top='off',
labelbottom='on',
right='off',
left='off',
labelleft='off',
labelsize=params['xtick.labelsize'])
if j == 0:
ax.set_ylabel(Ylabels[i],
size=params['font.size'],
labelpad=10)
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='on',
labelleft='on',
labelsize=params['xtick.labelsize'])
if i == enQOIs.shape[0] - 1 and j == 0:
ax.tick_params(axis='both',
which='both',
bottom='on',
top='off',
labelbottom='on',
right='off',
left='on',
labelleft='on',
labelsize=params['xtick.labelsize'])
if i != enQOIs.shape[0] - 1 and j != 0:
ax.tick_params(axis='both',
which='both',
bottom='off',
top='off',
labelbottom='off',
right='off',
left='off',
labelleft='off')
xMin, xMax = ax.get_xlim()
yMin, yMax = ax.get_ylim()
dx = (xMax - xMin) / 4
dy = (yMax - yMin) / 4
ax.set_xticks(
np.round([xMin + dx, xMax - dx], int(np.ceil(-np.log10(dx)))))
ax.set_yticks(
np.round([yMin + dy, yMax - dy], int(np.ceil(-np.log10(dy)))))
if not loadWeights: np.save(varsDir + '/wPerPair%i.npy' % step, wPerPair)
plt.tight_layout()
fig.subplots_adjust(left=0.21,
bottom=0.16,
right=0.99,
top=0.99,
hspace=0.,
wspace=0.)
plt.savefig('microMacroUQ_iterPF' + varsDir[-1] + '_%i.pdf' % step,
dpi=600)
#~ plt.show()
return wPerPair