Sparse Inversion with Iteratively Re-Weighted Least-Squares#

Least-squares inversion produces smooth models which may not be an accurate representation of the true model. Here we demonstrate the basics of inverting for sparse and/or blocky models. Here, we used the iteratively reweighted least-squares approach. For this tutorial, we focus on the following:

  • Defining the forward problem

  • Defining the inverse problem (data misfit, regularization, optimization)

  • Defining the paramters for the IRLS algorithm

  • Specifying directives for the inversion

  • Recovering a set of model parameters which explains the observations

import numpy as np
import matplotlib.pyplot as plt

from discretize import TensorMesh

from simpeg import (
    simulation,
    maps,
    data_misfit,
    directives,
    optimization,
    regularization,
    inverse_problem,
    inversion,
)

# sphinx_gallery_thumbnail_number = 3

Defining the Model and Mapping#

Here we generate a synthetic model and a mappig which goes from the model space to the row space of our linear operator.

nParam = 100  # Number of model paramters

# A 1D mesh is used to define the row-space of the linear operator.
mesh = TensorMesh([nParam])

# Creating the true model
true_model = np.zeros(mesh.nC)
true_model[mesh.cell_centers_x > 0.3] = 1.0
true_model[mesh.cell_centers_x > 0.45] = -0.5
true_model[mesh.cell_centers_x > 0.6] = 0

# Mapping from the model space to the row space of the linear operator
model_map = maps.IdentityMap(mesh)

# Plotting the true model
fig = plt.figure(figsize=(8, 5))
ax = fig.add_subplot(111)
ax.plot(mesh.cell_centers_x, true_model, "b-")
ax.set_ylim([-2, 2])
plot inv 2 inversion irls
(-2.0, 2.0)

Defining the Linear Operator#

Here we define the linear operator with dimensions (nData, nParam). In practive, you may have a problem-specific linear operator which you would like to construct or load here.

# Number of data observations (rows)
nData = 20

# Create the linear operator for the tutorial. The columns of the linear operator
# represents a set of decaying and oscillating functions.
jk = np.linspace(1.0, 60.0, nData)
p = -0.25
q = 0.25


def g(k):
    return np.exp(p * jk[k] * mesh.cell_centers_x) * np.cos(
        np.pi * q * jk[k] * mesh.cell_centers_x
    )


G = np.empty((nData, nParam))

for i in range(nData):
    G[i, :] = g(i)

# Plot the columns of G
fig = plt.figure(figsize=(8, 5))
ax = fig.add_subplot(111)
for i in range(G.shape[0]):
    ax.plot(G[i, :])

ax.set_title("Columns of matrix G")
Columns of matrix G
Text(0.5, 1.0, 'Columns of matrix G')

Defining the Simulation#

The simulation defines the relationship between the model parameters and predicted data.

sim = simulation.LinearSimulation(mesh, G=G, model_map=model_map)

Predict Synthetic Data#

Here, we use the true model to create synthetic data which we will subsequently invert.

# Standard deviation of Gaussian noise being added
std = 0.02
np.random.seed(1)

# Create a SimPEG data object
data_obj = sim.make_synthetic_data(true_model, noise_floor=std, add_noise=True)

Define the Inverse Problem#

The inverse problem is defined by 3 things:

  1. Data Misfit: a measure of how well our recovered model explains the field data

  2. Regularization: constraints placed on the recovered model and a priori information

  3. Optimization: the numerical approach used to solve the inverse problem

# Define the data misfit. Here the data misfit is the L2 norm of the weighted
# residual between the observed data and the data predicted for a given model.
# Within the data misfit, the residual between predicted and observed data are
# normalized by the data's standard deviation.
dmis = data_misfit.L2DataMisfit(simulation=sim, data=data_obj)

# Define the regularization (model objective function). Here, 'p' defines the
# the norm of the smallness term and 'q' defines the norm of the smoothness
# term.
reg = regularization.Sparse(mesh, mapping=model_map)
reg.reference_model = np.zeros(nParam)
p = 0.0
q = 0.0
reg.norms = [p, q]

# Define how the optimization problem is solved.
opt = optimization.ProjectedGNCG(
    maxIter=100, lower=-2.0, upper=2.0, maxIterLS=20, maxIterCG=30, tolCG=1e-4
)

# Here we define the inverse problem that is to be solved
inv_prob = inverse_problem.BaseInvProblem(dmis, reg, opt)
/home/vsts/work/1/s/simpeg/optimization.py:1535: FutureWarning:

InexactCG.tolCG has been deprecated, please use InexactCG.cg_atol. It will be removed in version 0.26.0 of SimPEG.

/home/vsts/work/1/s/simpeg/optimization.py:1061: FutureWarning:

InexactCG.maxIterCG has been deprecated, please use InexactCG.cg_maxiter. It will be removed in version 0.26.0 of SimPEG.

Define Inversion Directives#

Here we define any directiveas that are carried out during the inversion. This includes the cooling schedule for the trade-off parameter (beta), stopping criteria for the inversion and saving inversion results at each iteration.

# Add sensitivity weights but don't update at each beta
sensitivity_weights = directives.UpdateSensitivityWeights(every_iteration=False)

# Reach target misfit for L2 solution, then use IRLS until model stops changing.
IRLS = directives.UpdateIRLS(max_irls_iterations=40, f_min_change=1e-4)

# Defining a starting value for the trade-off parameter (beta) between the data
# misfit and the regularization.
starting_beta = directives.BetaEstimate_ByEig(beta0_ratio=1e0)

# Update the preconditionner
update_Jacobi = directives.UpdatePreconditioner()

# Save output at each iteration
saveDict = directives.SaveOutputEveryIteration(save_txt=False)

# Define the directives as a list
directives_list = [
    sensitivity_weights,
    IRLS,
    starting_beta,
    update_Jacobi,
    saveDict,
]
/home/vsts/work/1/s/simpeg/directives/_directives.py:1865: FutureWarning:

SaveEveryIteration.save_txt has been deprecated, please use SaveEveryIteration.on_disk. It will be removed in version 0.26.0 of SimPEG.

/home/vsts/work/1/s/simpeg/directives/_directives.py:1866: FutureWarning:

SaveEveryIteration.save_txt has been deprecated, please use SaveEveryIteration.on_disk. It will be removed in version 0.26.0 of SimPEG.

Setting a Starting Model and Running the Inversion#

To define the inversion object, we need to define the inversion problem and the set of directives. We can then run the inversion.

# Here we combine the inverse problem and the set of directives
inv = inversion.BaseInversion(inv_prob, directives_list)

# Starting model
starting_model = 1e-4 * np.ones(nParam)

# Run inversion
recovered_model = inv.run(starting_model)
Running inversion with SimPEG v0.24.1.dev37+gf5e5be547
================================================= Projected GNCG =================================================
  #     beta     phi_d     phi_m       f      |proj(x-g)-x|  LS   iter_CG   CG |Ax-b|/|b|  CG |Ax-b|   Comment
-----------------------------------------------------------------------------------------------------------------
   0  1.75e+06  3.68e+03  1.02e-09  3.68e+03                         0           inf          inf
   1  1.75e+06  1.87e+03  3.64e-04  2.51e+03    1.96e+01      0      18       1.16e-08     5.52e-05
   2  8.77e+05  1.29e+03  8.42e-04  2.03e+03    1.91e+01      0      17       6.84e-08     5.63e-05
   3  4.39e+05  7.66e+02  1.69e-03  1.51e+03    1.86e+01      0      18       8.02e-08     4.83e-05
   4  2.19e+05  3.87e+02  2.91e-03  1.02e+03    1.74e+01      0      21       6.37e-08     2.68e-05
   5  1.10e+05  1.69e+02  4.29e-03  6.39e+02    1.65e+01      0      25       2.61e-07     7.12e-05
   6  5.48e+04  6.66e+01  5.57e-03  3.72e+02    1.34e+01      0      30       7.87e-07     1.29e-04
   7  2.74e+04  2.57e+01  6.59e-03  2.06e+02    1.13e+01      0      30       3.86e-05     3.60e-03
   8  1.37e+04  1.08e+01  7.33e-03  1.11e+02    9.54e+00      0      30       8.20e-04     4.16e-02
Reached starting chifact with l2-norm regularization: Start IRLS steps...
irls_threshold 1.210108083793358
   9  1.37e+04  1.83e+01  9.02e-03  1.42e+02    1.45e+01      0      30       4.60e-04     1.44e-02
  10  1.37e+04  2.44e+01  9.87e-03  1.60e+02    1.34e+01      0      30       1.57e-04     3.29e-03
  11  1.08e+04  2.43e+01  1.10e-02  1.42e+02    3.18e+00      0      30       5.00e-04     5.55e-03
  12  8.46e+03  2.34e+01  1.19e-02  1.24e+02    3.37e+00      0      30       2.43e-04     2.37e-03
  13  6.79e+03  2.19e+01  1.26e-02  1.07e+02    4.59e+00      0      30       2.69e-04     2.40e-03
  14  6.79e+03  2.35e+01  1.24e-02  1.07e+02    9.68e+00      0      30       3.43e-05     3.48e-04
  15  5.43e+03  2.07e+01  1.26e-02  8.91e+01    6.36e+00      0      30       3.15e-05     3.54e-04
  16  5.43e+03  2.02e+01  1.17e-02  8.39e+01    8.08e+00      0      30       4.36e-05     3.93e-04
  17  5.43e+03  1.94e+01  1.08e-02  7.83e+01    8.54e+00      0      30       2.75e-05     2.61e-04
  18  5.43e+03  1.83e+01  9.85e-03  7.19e+01    8.80e+00      0      30       1.30e-05     1.28e-04
  19  5.43e+03  1.71e+01  8.85e-03  6.52e+01    9.20e+00      0      30       1.25e-05     1.31e-04
  20  8.61e+03  2.06e+01  7.17e-03  8.23e+01    1.68e+01      0      27       1.80e-06     7.10e-05
  21  8.61e+03  2.01e+01  6.44e-03  7.55e+01    1.13e+01      0      28       2.63e-06     4.42e-05
  22  8.61e+03  1.87e+01  5.64e-03  6.73e+01    1.16e+01      0      29       1.31e-06     2.33e-05
  23  8.61e+03  1.72e+01  4.89e-03  5.93e+01    1.18e+01      0      28       4.60e-06     9.06e-05
  24  1.36e+04  1.98e+01  3.99e-03  7.41e+01    1.91e+01      0      26       1.33e-06     9.46e-05
  25  1.36e+04  1.85e+01  3.44e-03  6.54e+01    1.26e+01      0      28       1.98e-06     5.48e-05
  26  1.36e+04  1.78e+01  2.97e-03  5.83e+01    1.39e+01      0      29       9.77e-07     3.24e-05
  27  2.12e+04  1.87e+01  2.35e-03  6.86e+01    1.98e+01      0      25       5.44e-07     6.14e-05
  28  2.12e+04  1.79e+01  1.99e-03  6.03e+01    1.48e+01      0      26       1.63e-06     7.87e-05
  29  3.31e+04  1.91e+01  1.57e-03  7.12e+01    1.92e+01      0      26       2.73e-07     4.16e-05
  30  3.31e+04  1.90e+01  1.32e-03  6.27e+01    1.56e+01      0      27       5.59e-07     3.58e-05
  31  3.31e+04  1.87e+01  1.08e-03  5.45e+01    1.85e+01      0      23       1.23e-06     8.33e-05
  32  3.31e+04  1.87e+01  8.95e-04  4.83e+01    1.47e+01      0      24       6.83e-07     3.72e-05
  33  3.31e+04  1.86e+01  7.41e-04  4.32e+01    1.46e+01      0      25       3.92e-07     2.91e-05
  34  3.31e+04  1.86e+01  6.19e-04  3.91e+01    1.46e+01      0      25       1.59e-06     7.52e-05
  35  3.31e+04  1.86e+01  5.16e-04  3.57e+01    1.45e+01      0      26       1.70e-06     8.31e-05
  36  3.31e+04  1.86e+01  4.30e-04  3.29e+01    1.60e+01      0      25       8.43e-07     4.08e-05
  37  3.31e+04  1.87e+01  3.60e-04  3.06e+01    1.47e+01      0      26       1.59e-06     8.26e-05
  38  3.31e+04  1.87e+01  3.01e-04  2.86e+01    1.48e+01      0      29       6.79e-07     3.62e-05
  39  3.31e+04  1.87e+01  2.52e-04  2.70e+01    1.48e+01      0      30       1.28e-06     6.97e-05
  40  3.31e+04  1.86e+01  2.10e-04  2.56e+01    1.49e+01      0      30       4.88e-06     2.72e-04
  41  3.31e+04  1.86e+01  1.76e-04  2.44e+01    1.51e+01      0      30       1.10e-05     6.27e-04
  42  3.31e+04  1.84e+01  1.47e-04  2.33e+01    1.50e+01      0      30       1.94e-05     1.14e-03
  43  3.31e+04  1.83e+01  1.23e-04  2.24e+01    1.50e+01      0      30       8.29e-05     6.79e-03
  44  3.31e+04  1.81e+01  1.03e-04  2.15e+01    1.49e+01      0      30       2.82e-04     1.71e-02
  45  3.31e+04  1.79e+01  8.43e-05  2.07e+01    1.50e+01      0      30       9.25e-06     5.71e-04
  46  5.17e+04  1.78e+01  6.70e-05  2.12e+01    1.73e+01      0      30       2.50e-06     7.20e-04
  47  8.07e+04  1.77e+01  5.47e-05  2.21e+01    1.72e+01      0      30       1.35e-07     3.78e-05
  48  1.26e+05  1.76e+01  4.47e-05  2.32e+01    1.71e+01      0      26       2.48e-07     6.93e-05
Reach maximum number of IRLS cycles: 40
------------------------- STOP! -------------------------
1 : |fc-fOld| = 1.8045e-01 <= tolF*(1+|f0|) = 3.6797e+02
1 : |xc-x_last| = 9.0348e-02 <= tolX*(1+|x0|) = 1.0010e-01
0 : |proj(x-g)-x|    = 1.7115e+01 <= tolG          = 1.0000e-01
0 : |proj(x-g)-x|    = 1.7115e+01 <= 1e3*eps       = 1.0000e-02
0 : maxIter   =     100    <= iter          =     48
------------------------- DONE! -------------------------

Plotting Results#

fig, ax = plt.subplots(1, 2, figsize=(12 * 1.2, 4 * 1.2))

# True versus recovered model
ax[0].plot(mesh.cell_centers_x, true_model, "k-")
ax[0].plot(mesh.cell_centers_x, inv_prob.l2model, "b-")
ax[0].plot(mesh.cell_centers_x, recovered_model, "r-")
ax[0].legend(("True Model", "Recovered L2 Model", "Recovered Sparse Model"))
ax[0].set_ylim([-2, 2])

# Observed versus predicted data
ax[1].plot(data_obj.dobs, "k-")
ax[1].plot(inv_prob.dpred, "ko")
ax[1].legend(("Observed Data", "Predicted Data"))

# Plot convergence
fig = plt.figure(figsize=(9, 5))
ax = fig.add_axes([0.2, 0.1, 0.7, 0.85])
ax.plot(saveDict.phi_d, "k", lw=2)

twin = ax.twinx()
twin.plot(saveDict.phi_m, "k--", lw=2)
ax.plot(
    np.r_[IRLS.metrics.start_irls_iter, IRLS.metrics.start_irls_iter],
    np.r_[0, np.max(saveDict.phi_d)],
    "k:",
)
ax.text(
    IRLS.metrics.start_irls_iter,
    0.0,
    "IRLS Start",
    va="bottom",
    ha="center",
    rotation="vertical",
    size=12,
    bbox={"facecolor": "white"},
)

ax.set_ylabel(r"$\phi_d$", size=16, rotation=0)
ax.set_xlabel("Iterations", size=14)
twin.set_ylabel(r"$\phi_m$", size=16, rotation=0)
  • plot inv 2 inversion irls
  • plot inv 2 inversion irls
Text(865.1527777777777, 0.5, '$\\phi_m$')

Total running time of the script: (0 minutes 27.827 seconds)

Estimated memory usage: 298 MB

Gallery generated by Sphinx-Gallery