def run_inversion_cg(
self,
maxIter=60,
m0=0.0,
mref=0.0,
percentage=5,
floor=0.1,
chifact=1,
beta0_ratio=1.0,
coolingFactor=1,
coolingRate=1,
alpha_s=1.0,
alpha_x=1.0,
use_target=False,
):
survey, prob = self.get_problem_survey()
survey.eps = percentage
survey.std = floor
survey.dobs = self.data.copy()
self.uncertainty = percentage * abs(survey.dobs) * 0.01 + floor
m0 = np.ones(self.M) * m0
mref = np.ones(self.M) * mref
reg = Regularization.Tikhonov(
self.mesh, alpha_s=alpha_s, alpha_x=alpha_x, mref=mref
)
dmis = DataMisfit.l2_DataMisfit(survey)
dmis.W = 1.0 / self.uncertainty
opt = Optimization.InexactGaussNewton(maxIter=maxIter, maxIterCG=20)
opt.remember("xc")
opt.tolG = 1e-10
opt.eps = 1e-10
invProb = InvProblem.BaseInvProblem(dmis, reg, opt)
save = Directives.SaveOutputEveryIteration()
beta_schedule = Directives.BetaSchedule(
coolingFactor=coolingFactor, coolingRate=coolingRate
)
target = Directives.TargetMisfit(chifact=chifact)
if use_target:
directives = [
Directives.BetaEstimate_ByEig(beta0_ratio=beta0_ratio),
beta_schedule,
target,
save,
]
else:
directives = [
Directives.BetaEstimate_ByEig(beta0_ratio=beta0_ratio),
beta_schedule,
save,
]
inv = Inversion.BaseInversion(invProb, directiveList=directives)
mopt = inv.run(m0)
model = opt.recall("xc")
model.append(mopt)
pred = []
for m in model:
pred.append(survey.dpred(m))
return model, pred, save
# Specify how the initial beta is found
# if input_dict["inversion_type"].lower() in ['mvi', 'mvis']:
betaest = Directives.BetaEstimate_ByEig(beta0_ratio=1e+1)
# Pre-conditioner
update_Jacobi = Directives.UpdatePreconditioner()
IRLS = Directives.Update_IRLS(f_min_change=1e-3,
minGNiter=1,
beta_tol=0.25,
maxIRLSiter=max_IRLS_iter,
chifact_target=target_chi,
betaSearch=False)
# Save model
saveDict = Directives.SaveOutputEveryIteration(save_txt=False)
saveIt = Directives.SaveUBCModelEveryIteration(
mapping=activeCellsMap,
fileName=outDir + input_dict["inversion_type"].lower() + "_C",
vector=input_dict["inversion_type"].lower()[0:3] == 'mvi')
# Put all the parts together
inv = Inversion.BaseInversion(
invProb, directiveList=[saveIt, saveDict, betaest, IRLS, update_Jacobi])
# SimPEG reports half phi_d, so we scale to matrch
print("Start Inversion\nTarget Misfit: %.2e (%.0f data with chifact = %g)" %
(0.5 * target_chi * len(survey.std), len(survey.std), target_chi))
# Run the inversion
mrec = inv.run(mstart)
problem.pair(survey)
problem.Solver = Solver
regmap = Maps.IdentityMap(nP=m0.size)
survey.std = perc
survey.eps = floor
survey.dobs = dobs
dmisfit = DataMisfit.l2_DataMisfit(survey)
reg = Regularization.Simple(mesh, mapping=regmap, indActive=~airind)
opt = Optimization.InexactGaussNewton(maxIter=20)
invProb = InvProblem.BaseInvProblem(dmisfit, reg, opt)
# Create an inversion object
beta = Directives.BetaSchedule(coolingFactor=5, coolingRate=2)
betaest = Directives.BetaEstimate_ByEig(beta0_ratio=1e0)
save = Directives.SaveOutputEveryIteration()
save_model = Directives.SaveModelEveryIteration()
target = Directives.TargetMisfit()
inv = Inversion.BaseInversion(
invProb, directiveList=[beta, betaest, save, save_model, target])
reg.alpha_s = 1e-2
reg.alpha_x = 1.
reg.alpha_y = 1.
reg.alpha_z = 1.
problem.counter = opt.counter = Utils.Counter()
opt.LSshorten = 0.5
mopt = inv.run(m0)
sigopt = mapping * mopt
np.save("sigest", sigopt)
def run(plotIt=True):
# Define the inducing field parameter
H0 = (50000, 90, 0)
# Create a mesh
dx = 5.
hxind = [(dx, 5, -1.3), (dx, 10), (dx, 5, 1.3)]
hyind = [(dx, 5, -1.3), (dx, 10), (dx, 5, 1.3)]
hzind = [(dx, 5, -1.3), (dx, 10)]
mesh = Mesh.TensorMesh([hxind, hyind, hzind], 'CCC')
# Get index of the center
midx = int(mesh.nCx / 2)
midy = int(mesh.nCy / 2)
# Lets create a simple Gaussian topo and set the active cells
[xx, yy] = np.meshgrid(mesh.vectorNx, mesh.vectorNy)
zz = -np.exp((xx**2 + yy**2) / 75**2) + mesh.vectorNz[-1]
# We would usually load a topofile
topo = np.c_[Utils.mkvc(xx), Utils.mkvc(yy), Utils.mkvc(zz)]
# Go from topo to array of indices of active cells
actv = Utils.surface2ind_topo(mesh, topo, 'N')
actv = np.where(actv)[0]
nC = len(actv)
# Create and array of observation points
xr = np.linspace(-20., 20., 20)
yr = np.linspace(-20., 20., 20)
X, Y = np.meshgrid(xr, yr)
# Move the observation points 5m above the topo
Z = -np.exp((X**2 + Y**2) / 75**2) + mesh.vectorNz[-1] + 5.
# Create a MAGsurvey
rxLoc = np.c_[Utils.mkvc(X.T), Utils.mkvc(Y.T), Utils.mkvc(Z.T)]
rxLoc = PF.BaseMag.RxObs(rxLoc)
srcField = PF.BaseMag.SrcField([rxLoc], param=H0)
survey = PF.BaseMag.LinearSurvey(srcField)
# We can now create a susceptibility model and generate data
# Here a simple block in half-space
model = np.zeros((mesh.nCx, mesh.nCy, mesh.nCz))
model[(midx - 2):(midx + 2), (midy - 2):(midy + 2), -6:-2] = 0.02
model = Utils.mkvc(model)
model = model[actv]
# Create active map to go from reduce set to full
actvMap = Maps.InjectActiveCells(mesh, actv, -100)
# Create reduced identity map
idenMap = Maps.IdentityMap(nP=nC)
# Create the forward model operator
prob = PF.Magnetics.MagneticIntegral(mesh, chiMap=idenMap, actInd=actv)
# Pair the survey and problem
survey.pair(prob)
# Compute linear forward operator and compute some data
d = prob.fields(model)
# Add noise and uncertainties
# We add some random Gaussian noise (1nT)
data = d + np.random.randn(len(d))
wd = np.ones(len(data)) * 1. # Assign flat uncertainties
survey.dobs = data
survey.std = wd
survey.mtrue = model
# Create sensitivity weights from our linear forward operator
rxLoc = survey.srcField.rxList[0].locs
wr = np.sum(prob.G**2., axis=0)**0.5
wr = (wr / np.max(wr))
# Create a regularization
reg = Regularization.Sparse(mesh, indActive=actv, mapping=idenMap)
reg.cell_weights = wr
reg.mref = np.zeros(nC)
reg.norms = np.c_[0, 0, 0, 0]
# reg.eps_p, reg.eps_q = 1e-0, 1e-0
# Data misfit function
dmis = DataMisfit.l2_DataMisfit(survey)
dmis.W = 1 / wd
# Add directives to the inversion
opt = Optimization.ProjectedGNCG(maxIter=100,
lower=0.,
upper=1.,
maxIterLS=20,
maxIterCG=20,
tolCG=1e-3)
invProb = InvProblem.BaseInvProblem(dmis, reg, opt)
betaest = Directives.BetaEstimate_ByEig(beta0_ratio=1e-1)
# Here is where the norms are applied
# Use pick a threshold parameter empirically based on the distribution of
# model parameters
IRLS = Directives.Update_IRLS(f_min_change=1e-4, maxIRLSiter=40)
saveDict = Directives.SaveOutputEveryIteration(save_txt=False)
update_Jacobi = Directives.UpdatePreconditioner()
inv = Inversion.BaseInversion(
invProb, directiveList=[IRLS, betaest, update_Jacobi, saveDict])
# Run the inversion
m0 = np.ones(nC) * 1e-4 # Starting model
mrec = inv.run(m0)
if plotIt:
# Here is the recovered susceptibility model
ypanel = midx
zpanel = -5
m_l2 = actvMap * invProb.l2model
m_l2[m_l2 == -100] = np.nan
m_lp = actvMap * mrec
m_lp[m_lp == -100] = np.nan
m_true = actvMap * model
m_true[m_true == -100] = np.nan
# Plot the data
Utils.PlotUtils.plot2Ddata(rxLoc, d)
plt.figure()
# Plot L2 model
ax = plt.subplot(321)
mesh.plotSlice(m_l2,
ax=ax,
normal='Z',
ind=zpanel,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCy[ypanel], mesh.vectorCCy[ypanel]]),
color='w')
plt.title('Plan l2-model.')
plt.gca().set_aspect('equal')
plt.ylabel('y')
ax.xaxis.set_visible(False)
plt.gca().set_aspect('equal', adjustable='box')
# Vertica section
ax = plt.subplot(322)
mesh.plotSlice(m_l2,
ax=ax,
normal='Y',
ind=midx,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCz[zpanel], mesh.vectorCCz[zpanel]]),
color='w')
plt.title('E-W l2-model.')
plt.gca().set_aspect('equal')
ax.xaxis.set_visible(False)
plt.ylabel('z')
plt.gca().set_aspect('equal', adjustable='box')
# Plot Lp model
ax = plt.subplot(323)
mesh.plotSlice(m_lp,
ax=ax,
normal='Z',
ind=zpanel,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCy[ypanel], mesh.vectorCCy[ypanel]]),
color='w')
plt.title('Plan lp-model.')
plt.gca().set_aspect('equal')
ax.xaxis.set_visible(False)
plt.ylabel('y')
plt.gca().set_aspect('equal', adjustable='box')
# Vertical section
ax = plt.subplot(324)
mesh.plotSlice(m_lp,
ax=ax,
normal='Y',
ind=midx,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCz[zpanel], mesh.vectorCCz[zpanel]]),
color='w')
plt.title('E-W lp-model.')
plt.gca().set_aspect('equal')
ax.xaxis.set_visible(False)
plt.ylabel('z')
plt.gca().set_aspect('equal', adjustable='box')
# Plot True model
ax = plt.subplot(325)
mesh.plotSlice(m_true,
ax=ax,
normal='Z',
ind=zpanel,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCy[ypanel], mesh.vectorCCy[ypanel]]),
color='w')
plt.title('Plan true model.')
plt.gca().set_aspect('equal')
plt.xlabel('x')
plt.ylabel('y')
plt.gca().set_aspect('equal', adjustable='box')
# Vertical section
ax = plt.subplot(326)
mesh.plotSlice(m_true,
ax=ax,
normal='Y',
ind=midx,
grid=True,
clim=(model.min(), model.max()))
plt.plot(([mesh.vectorCCx[0], mesh.vectorCCx[-1]]),
([mesh.vectorCCz[zpanel], mesh.vectorCCz[zpanel]]),
color='w')
plt.title('E-W true model.')
plt.gca().set_aspect('equal')
plt.xlabel('x')
plt.ylabel('z')
plt.gca().set_aspect('equal', adjustable='box')
# Plot convergence curves
fig, axs = plt.figure(), plt.subplot()
axs.plot(saveDict.phi_d, 'k', lw=2)
axs.plot(np.r_[IRLS.iterStart, IRLS.iterStart],
np.r_[0, np.max(saveDict.phi_d)], 'k:')
twin = axs.twinx()
twin.plot(saveDict.phi_m, 'k--', lw=2)
axs.text(IRLS.iterStart,
0,
'IRLS Steps',
va='bottom',
ha='center',
rotation='vertical',
size=12,
bbox={'facecolor': 'white'})
axs.set_ylabel('$\phi_d$', size=16, rotation=0)
axs.set_xlabel('Iterations', size=14)
twin.set_ylabel('$\phi_m$', size=16, rotation=0)
def run_inversion(
self,
maxIter=60,
m0=0.0,
mref=0.0,
percentage=5,
floor=0.1,
chifact=1,
beta0_ratio=1.0,
coolingFactor=1,
n_iter_per_beta=1,
alpha_s=1.0,
alpha_x=1.0,
alpha_z=1.0,
use_target=False,
use_tikhonov=True,
use_irls=False,
p_s=2,
p_x=2,
p_y=2,
p_z=2,
beta_start=None,
):
self.uncertainty = percentage * abs(self.survey.dobs) * 0.01 + floor
m0 = np.ones(self.mesh.nC) * m0
mref = np.ones(self.mesh.nC) * mref
if ~use_tikhonov:
reg = Regularization.Sparse(
self.mesh,
alpha_s=alpha_s,
alpha_x=alpha_x,
alpha_y=alpha_z,
mref=mref,
mapping=Maps.IdentityMap(self.mesh),
cell_weights=self.mesh.vol,
)
else:
reg = Regularization.Tikhonov(
self.mesh,
alpha_s=alpha_s,
alpha_x=alpha_x,
alpha_y=alpha_z,
mref=mref,
mapping=Maps.IdentityMap(self.mesh),
)
dmis = DataMisfit.l2_DataMisfit(self.survey)
dmis.W = 1.0 / self.uncertainty
opt = Optimization.ProjectedGNCG(maxIter=maxIter, maxIterCG=20)
opt.lower = 0.0
opt.remember("xc")
opt.tolG = 1e-10
opt.eps = 1e-10
invProb = InvProblem.BaseInvProblem(dmis, reg, opt)
save = Directives.SaveOutputEveryIteration()
beta_schedule = Directives.BetaSchedule(coolingFactor=coolingFactor,
coolingRate=n_iter_per_beta)
if use_irls:
IRLS = Directives.Update_IRLS(
f_min_change=1e-4,
minGNiter=1,
silent=False,
maxIRLSiter=40,
beta_tol=5e-1,
coolEpsFact=1.3,
chifact_start=chifact,
)
if beta_start is None:
directives = [
Directives.BetaEstimate_ByEig(beta0_ratio=beta0_ratio),
IRLS,
save,
]
else:
directives = [IRLS, save]
invProb.beta = beta_start
reg.norms = np.c_[p_s, p_x, p_z, 2]
else:
target = Directives.TargetMisfit(chifact=chifact)
directives = [
Directives.BetaEstimate_ByEig(beta0_ratio=beta0_ratio),
beta_schedule,
save,
]
if use_target:
directives.append(target)
inv = Inversion.BaseInversion(invProb, directiveList=directives)
mopt = inv.run(m0)
model = opt.recall("xc")
model.append(mopt)
pred = []
for m in model:
pred.append(self.survey.dpred(m))
return model, pred, save
def run(N=100, plotIt=True):
np.random.seed(1)
std_noise = 1e-2
mesh = Mesh.TensorMesh([N])
m0 = np.ones(mesh.nC) * 1e-4
mref = np.zeros(mesh.nC)
nk = 20
jk = np.linspace(1., 60., nk)
p = -0.25
q = 0.25
def g(k):
return (
np.exp(p*jk[k]*mesh.vectorCCx) *
np.cos(np.pi*q*jk[k]*mesh.vectorCCx)
)
G = np.empty((nk, mesh.nC))
for i in range(nk):
G[i, :] = g(i)
mtrue = np.zeros(mesh.nC)
mtrue[mesh.vectorCCx > 0.3] = 1.
mtrue[mesh.vectorCCx > 0.45] = -0.5
mtrue[mesh.vectorCCx > 0.6] = 0
prob = Problem.LinearProblem(mesh, G=G)
survey = Survey.LinearSurvey()
survey.pair(prob)
survey.dobs = prob.fields(mtrue) + std_noise * np.random.randn(nk)
wd = np.ones(nk) * std_noise
# Distance weighting
wr = np.sum(prob.getJ(m0)**2., axis=0)**0.5
wr = wr/np.max(wr)
dmis = DataMisfit.l2_DataMisfit(survey)
dmis.W = 1./wd
betaest = Directives.BetaEstimate_ByEig(beta0_ratio=1e0)
# Creat reduced identity map
idenMap = Maps.IdentityMap(nP=mesh.nC)
reg = Regularization.Sparse(mesh, mapping=idenMap)
reg.mref = mref
reg.cell_weights = wr
reg.norms = np.c_[0., 0., 2., 2.]
reg.mref = np.zeros(mesh.nC)
opt = Optimization.ProjectedGNCG(
maxIter=100, lower=-2., upper=2.,
maxIterLS=20, maxIterCG=10, tolCG=1e-3
)
invProb = InvProblem.BaseInvProblem(dmis, reg, opt)
update_Jacobi = Directives.UpdatePreconditioner()
# Set the IRLS directive, penalize the lowest 25 percentile of model values
# Start with an l2-l2, then switch to lp-norms
IRLS = Directives.Update_IRLS(
maxIRLSiter=40, minGNiter=1, f_min_change=1e-4)
saveDict = Directives.SaveOutputEveryIteration(save_txt=False)
inv = Inversion.BaseInversion(
invProb,
directiveList=[IRLS, betaest, update_Jacobi, saveDict]
)
# Run inversion
mrec = inv.run(m0)
print("Final misfit:" + str(invProb.dmisfit(mrec)))
if plotIt:
fig, axes = plt.subplots(2, 2, figsize=(12*1.2, 8*1.2))
for i in range(prob.G.shape[0]):
axes[0, 0].plot(prob.G[i, :])
axes[0, 0].set_title('Columns of matrix G')
axes[0, 1].plot(mesh.vectorCCx, mtrue, 'b-')
axes[0, 1].plot(mesh.vectorCCx, invProb.l2model, 'r-')
# axes[0, 1].legend(('True Model', 'Recovered Model'))
axes[0, 1].set_ylim(-1.0, 1.25)
axes[0, 1].plot(mesh.vectorCCx, mrec, 'k-', lw=2)
axes[0, 1].legend(
(
'True Model',
'Smooth l2-l2',
'Sparse norms: {0}'.format(*reg.norms)
),
fontsize=12
)
axes[1, 1].plot(saveDict.phi_d, 'k', lw=2)
twin = axes[1, 1].twinx()
twin.plot(saveDict.phi_m, 'k--', lw=2)
axes[1, 1].plot(
np.r_[IRLS.iterStart, IRLS.iterStart],
np.r_[0, np.max(saveDict.phi_d)], 'k:'
)
axes[1, 1].text(
IRLS.iterStart, 0.,
'IRLS Start', va='bottom', ha='center',
rotation='vertical', size=12,
bbox={'facecolor': 'white'}
)
axes[1, 1].set_ylabel('$\phi_d$', size=16, rotation=0)
axes[1, 1].set_xlabel('Iterations', size=14)
axes[1, 0].axis('off')
twin.set_ylabel('$\phi_m$', size=16, rotation=0)
return prob, survey, mesh, mrec
