A Brief Exploration of open-source gradient-based numerical optimization Python libraries for full-waveform inversion.
This repository contains the implementation to reproduce the numerical experiments of the Seismica paper: A Brief Exploration of open-source gradient-based numerical optimization Python libraries for full-waveform inversion.
👉 To find your way around the code, the mymakefile file is the recommended starting point.
Along with the code a Dockerfile from which you can build a docker image for our Python application is provided.
You can build the docker image, named myapp, with the following command
docker build --tag myapp .… then wait for docker to build an image for our app (it might tight a take for downloading and update it the packages list).
Make sure Docker is installed on your machine (www.docker.com).
To run all the experiments used in the paper from the docker image use
docker run --rm -ti myapp make -f mymakefileThis execute the make command with the specified Makefile (mymakefile), which invoke python scripts for forward modeling and fwi. When make -f mymakefile is run, forward will be executed first, followed by fwi. You also can run fwi with each one of the frameworks individually, i.e., docker run --rm -ti myapp make -f mymakefile scipy, will be execute fwi using scipy, forward will be executed first if it has not been executed yet
The command above should outputs:
python3 generate_shot_data.py marmousi2
Solver parameters are being substituted with the following new values:
Key: shotfile_path Value: ./marmousi2/shots/
Key: parfile_path Value: ./marmousi2/parameters_hdf5/
Key: t0 Value: 0.0
Key: tn Value: 5000.0
Key: dt Value: 4.0
Key: f0 Value: 0.004
Key: model_name Value: marmousi2
Key: nbl Value: 50
Key: space_order Value: 8
Key: dtype Value: float32
Running Forward modeling ...
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
/opt/conda/lib/python3.10/site-packages/segyio/utils.py:18: RuntimeWarning: Implicit conversion to contiguous array
warnings.warn(msg, RuntimeWarning)
/opt/conda/lib/python3.10/site-packages/segyio/utils.py:18: RuntimeWarning: Implicit conversion to contiguous array
warnings.warn(msg, RuntimeWarning)
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
/opt/conda/lib/python3.10/site-packages/segyio/utils.py:18: RuntimeWarning: Implicit conversion to contiguous array
warnings.warn(msg, RuntimeWarning)
/opt/conda/lib/python3.10/site-packages/segyio/utils.py:18: RuntimeWarning: Implicit conversion to contiguous array
warnings.warn(msg, RuntimeWarning)
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Shot with time interval of 4.427000045776367 ms
Forward modeling took :- 00:00:11
Successfully generated 16 shots
python3 inversion_script.py shot_control_inversion_scipy || true
Running fwi ...
Cost_fcn eval took 9.91 sec - Cost_fcn= 9.732E+06
RUNNING THE L-BFGS-B CODE
* * *
Machine precision = 2.220D-16
N = 37488 M = 10
At iterate 0 f= 9.73177D+06 |proj g|= 4.73626D-01
Cost_fcn eval took 6.33 sec - Cost_fcn= 1.040E+08
Cost_fcn eval took 7.10 sec - Cost_fcn= 9.728E+06
Cost_fcn eval took 6.32 sec - Cost_fcn= 6.844E+07
Cost_fcn eval took 7.06 sec - Cost_fcn= 9.726E+06
Cost_fcn eval took 6.26 sec - Cost_fcn= 9.726E+06
Cost_fcn eval took 6.99 sec - Cost_fcn= 9.726E+06
At iterate 1 f= 9.72588D+06 |proj g|= 4.73617D-01
Cost_fcn eval took 7.01 sec - Cost_fcn= 7.308E+06
At iterate 2 f= 7.30833D+06 |proj g|= 4.52685D-01
Cost_fcn eval took 6.34 sec - Cost_fcn= 6.161E+06
At iterate 3 f= 6.16085D+06 |proj g|= 4.36160D-01
Cost_fcn eval took 7.29 sec - Cost_fcn= 3.732E+06
At iterate 4 f= 3.73238D+06 |proj g|= 4.31804D-01
Cost_fcn eval took 6.40 sec - Cost_fcn= 3.485E+06
At iterate 5 f= 3.48450D+06 |proj g|= 4.81889D-01
Cost_fcn eval took 6.70 sec - Cost_fcn= 2.548E+06
At iterate 6 f= 2.54849D+06 |proj g|= 4.80791D-01
Cost_fcn eval took 7.02 sec - Cost_fcn= 1.881E+06
At iterate 7 f= 1.88088D+06 |proj g|= 4.76795D-01
Cost_fcn eval took 6.46 sec - Cost_fcn= 1.485E+06
At iterate 8 f= 1.48542D+06 |proj g|= 4.66909D-01
Cost_fcn eval took 6.91 sec - Cost_fcn= 1.049E+06
At iterate 9 f= 1.04931D+06 |proj g|= 4.67145D-01
Cost_fcn eval took 6.60 sec - Cost_fcn= 8.750E+05
At iterate 10 f= 8.75026D+05 |proj g|= 4.62418D-01
Cost_fcn eval took 6.96 sec - Cost_fcn= 7.302E+05
At iterate 11 f= 7.30162D+05 |proj g|= 4.14073D-01
Cost_fcn eval took 6.88 sec - Cost_fcn= 6.127E+05
At iterate 12 f= 6.12679D+05 |proj g|= 3.90788D-01
Cost_fcn eval took 6.81 sec - Cost_fcn= 5.161E+05
At iterate 13 f= 5.16131D+05 |proj g|= 3.65810D-01
Cost_fcn eval took 7.37 sec - Cost_fcn= 3.746E+05
At iterate 14 f= 3.74626D+05 |proj g|= 4.58039D-01
Cost_fcn eval took 6.45 sec - Cost_fcn= 3.195E+05
At iterate 15 f= 3.19487D+05 |proj g|= 3.77208D-01
Cost_fcn eval took 7.17 sec - Cost_fcn= 2.913E+05
At iterate 16 f= 2.91265D+05 |proj g|= 3.50149D-01
Cost_fcn eval took 6.32 sec - Cost_fcn= 2.743E+05
At iterate 17 f= 2.74300D+05 |proj g|= 2.36171D-01
Cost_fcn eval took 6.98 sec - Cost_fcn= 2.499E+05
At iterate 18 f= 2.49862D+05 |proj g|= 3.14808D-01
Cost_fcn eval took 6.66 sec - Cost_fcn= 2.069E+05
At iterate 19 f= 2.06865D+05 |proj g|= 3.36018D-01
Cost_fcn eval took 6.42 sec - Cost_fcn= 1.540E+05
At iterate 20 f= 1.53984D+05 |proj g|= 3.46861D-01
* * *
Tit = total number of iterations
Tnf = total number of function evaluations
Tnint = total number of segments explored during Cauchy searches
Skip = number of BFGS updates skipped
Nact = number of active bounds at final generalized Cauchy point
Projg = norm of the final projected gradient
F = final function value
* * *
N Tit Tnf Tnint Skip Nact Projg F
37488 20 26 22796 0 72 3.469D-01 1.540D+05
F = 153983.64062500000
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT
Iterative inversion took :- 00:03:35
python3 fwi_marmousi2_pyrol_trillinos_daskcluster.py
Running fwi ...
L-Secant-B Line-Search Method (Type B, Bound Constraints)
iter value gnorm snorm LSpar #fval #grad #proj iterCG flagCG
0 9.731758e+06 5.290107e+01 --- 1.000000e+00 1 1 2 --- ---
1 8.409304e+06 2.966844e+01 4.831089e-01 7.812500e-03 9 2 11 0 0
2 7.392617e+06 3.098177e+01 1.836908e-01 1.000000e+00 10 3 22 1 0
3 6.072640e+06 3.763823e+01 3.350746e-01 1.000000e+00 11 4 27 1 0
4 4.824638e+06 4.469295e+01 4.384678e-01 1.000000e+00 12 5 32 1 2
5 4.370995e+06 2.716177e+01 1.252939e-01 1.000000e+00 13 6 37 1 2
6 4.128030e+06 3.081551e+01 8.103123e-02 1.000000e+00 14 7 42 1 2
7 2.719162e+06 2.643041e+01 5.977432e-01 1.000000e+00 15 8 48 1 2
8 2.410463e+06 2.931974e+01 1.335843e-01 1.000000e+00 16 9 52 1 2
9 2.268405e+06 2.334232e+01 6.713458e-02 1.000000e+00 17 10 57 1 2
10 2.087515e+06 2.255981e+01 9.248706e-02 1.000000e+00 18 11 62 1 2
11 1.547560e+06 3.068538e+01 3.342841e-01 1.000000e+00 19 12 68 1 2
12 1.466858e+06 1.777505e+01 4.734497e-02 1.000000e+00 20 13 72 1 2
13 1.406804e+06 1.850941e+01 4.150382e-02 1.000000e+00 21 14 77 1 2
14 1.118984e+06 2.296888e+01 2.381057e-01 1.000000e+00 22 15 83 1 2
15 1.072215e+06 2.146037e+01 4.733508e-02 1.000000e+00 23 16 87 1 2
16 1.040634e+06 1.485732e+01 2.652437e-02 1.000000e+00 24 17 92 1 2
17 1.013475e+06 1.468449e+01 2.328420e-02 1.000000e+00 25 18 97 1 2
18 8.250006e+05 1.873156e+01 1.802607e-01 1.000000e+00 26 19 103 1 2
19 8.022595e+05 1.683758e+01 3.220533e-02 1.000000e+00 27 20 107 1 2
20 7.755471e+05 1.198507e+01 2.535021e-02 1.000000e+00 28 21 112 1 2
Optimization Terminated with Status: Iteration Limit Exceeded
python3 inversion_script.py shot_control_inversion_sotb
Running fwi ...
Cost_fcn eval took 9.50 sec - Cost_fcn= 9.732E+06
Cost_fcn eval took 6.70 sec - Cost_fcn= 1.040E+08
Cost_fcn eval took 6.60 sec - Cost_fcn= 9.994E+07
Cost_fcn eval took 6.59 sec - Cost_fcn= 9.161E+07
Cost_fcn eval took 6.99 sec - Cost_fcn= 7.360E+07
Cost_fcn eval took 7.03 sec - Cost_fcn= 5.343E+07
Cost_fcn eval took 6.67 sec - Cost_fcn= 4.047E+07
Cost_fcn eval took 6.82 sec - Cost_fcn= 2.542E+07
Cost_fcn eval took 7.02 sec - Cost_fcn= 1.440E+07
Cost_fcn eval took 6.95 sec - Cost_fcn= 8.016E+06
Cost_fcn eval took 6.81 sec - Cost_fcn= 8.005E+06
Cost_fcn eval took 6.70 sec - Cost_fcn= 7.914E+06
Cost_fcn eval took 6.84 sec - Cost_fcn= 7.309E+06
Cost_fcn eval took 6.94 sec - Cost_fcn= 5.941E+06
Cost_fcn eval took 6.90 sec - Cost_fcn= 5.692E+06
Cost_fcn eval took 6.74 sec - Cost_fcn= 6.512E+06
Cost_fcn eval took 6.21 sec - Cost_fcn= 4.761E+06
Cost_fcn eval took 6.51 sec - Cost_fcn= 4.121E+06
Cost_fcn eval took 6.61 sec - Cost_fcn= 3.400E+06
Cost_fcn eval took 6.73 sec - Cost_fcn= 2.596E+06
Cost_fcn eval took 6.72 sec - Cost_fcn= 1.733E+06
Cost_fcn eval took 6.46 sec - Cost_fcn= 1.349E+06
Cost_fcn eval took 6.22 sec - Cost_fcn= 1.288E+06
Cost_fcn eval took 6.50 sec - Cost_fcn= 1.183E+06
Cost_fcn eval took 6.58 sec - Cost_fcn= 1.173E+06
Cost_fcn eval took 6.71 sec - Cost_fcn= 1.162E+06
Cost_fcn eval took 7.06 sec - Cost_fcn= 1.209E+06
Cost_fcn eval took 7.02 sec - Cost_fcn= 7.589E+05
Cost_fcn eval took 7.06 sec - Cost_fcn= 7.129E+05
Cost_fcn eval took 7.02 sec - Cost_fcn= 4.281E+05
Cost_fcn eval took 7.20 sec - Cost_fcn= 5.054E+05
Cost_fcn eval took 6.42 sec - Cost_fcn= 3.990E+05
Cost_fcn eval took 7.31 sec - Cost_fcn= 3.794E+05
Cost_fcn eval took 6.96 sec - Cost_fcn= 3.601E+05
Cost_fcn eval took 6.75 sec - Cost_fcn= 3.378E+05
Iterative inversion took :- 00:04:33
END OF TEST
FINAL iterate is : [0.44444445 0.44444445 0.44444445 ... 0.0743812 0.07149379 0.06978292]
See the convergence history in iterate_LBFGS.dat
python3 inversion_script.py shot_control_inversion_nlopt
Running fwi ...
Iteration Function Val norm(g)
Cost_fcn eval took 9.38 sec - Cost_fcn= 9.732E+06
1 9.73176e+06 1.84245e+02
Cost_fcn eval took 6.61 sec - Cost_fcn= 2.303E+07
2 2.30265e+07 4.06681e+02
Cost_fcn eval took 6.85 sec - Cost_fcn= 9.227E+06
3 9.22650e+06 1.89238e+02
Cost_fcn eval took 6.67 sec - Cost_fcn= 3.064E+07
4 3.06386e+07 4.38373e+02
Cost_fcn eval took 6.37 sec - Cost_fcn= 9.338E+06
5 9.33806e+06 1.61961e+02
Cost_fcn eval took 6.00 sec - Cost_fcn= 9.225E+06
6 9.22467e+06 1.89086e+02
Cost_fcn eval took 6.70 sec - Cost_fcn= 6.797E+06
7 6.79651e+06 1.61567e+02
Cost_fcn eval took 6.65 sec - Cost_fcn= 6.406E+06
8 6.40554e+06 2.18455e+02
Cost_fcn eval took 6.96 sec - Cost_fcn= 5.882E+06
9 5.88231e+06 1.77999e+02
Cost_fcn eval took 7.08 sec - Cost_fcn= 4.255E+06
10 4.25493e+06 1.02999e+02
Cost_fcn eval took 7.13 sec - Cost_fcn= 4.240E+06
11 4.24012e+06 1.02416e+02
Cost_fcn eval took 7.12 sec - Cost_fcn= 3.858E+06
12 3.85843e+06 9.05354e+01
Cost_fcn eval took 7.57 sec - Cost_fcn= 3.697E+06
13 3.69661e+06 8.60576e+01
Cost_fcn eval took 6.96 sec - Cost_fcn= 3.626E+06
14 3.62589e+06 8.45011e+01
Cost_fcn eval took 7.57 sec - Cost_fcn= 3.620E+06
15 3.62012e+06 8.43682e+01
Cost_fcn eval took 8.44 sec - Cost_fcn= 3.482E+06
16 3.48191e+06 8.17255e+01
Cost_fcn eval took 6.92 sec - Cost_fcn= 3.410E+06
17 3.41048e+06 8.03531e+01
Cost_fcn eval took 7.29 sec - Cost_fcn= 3.231E+06
18 3.23117e+06 7.72018e+01
Cost_fcn eval took 7.24 sec - Cost_fcn= 3.096E+06
19 3.09552e+06 7.49269e+01
Cost_fcn eval took 6.55 sec - Cost_fcn= 2.832E+06
20 2.83249e+06 7.04207e+01
Cost_fcn eval took 6.78 sec - Cost_fcn= 2.697E+06
21 2.69723e+06 6.81168e+01
Cost_fcn eval took 7.23 sec - Cost_fcn= 2.650E+06
22 2.64997e+06 6.72629e+01
Cost_fcn eval took 7.22 sec - Cost_fcn= 2.593E+06
23 2.59312e+06 6.62590e+01
Cost_fcn eval took 7.05 sec - Cost_fcn= 2.557E+06
24 2.55732e+06 6.57444e+01
Cost_fcn eval took 6.81 sec - Cost_fcn= 2.515E+06
25 2.51548e+06 6.49666e+01
Cost_fcn eval took 7.38 sec - Cost_fcn= 2.400E+06
26 2.40029e+06 6.36432e+01
Cost_fcn eval took 8.37 sec - Cost_fcn= 2.346E+06
27 2.34640e+06 6.24542e+01
Cost_fcn eval took 6.92 sec - Cost_fcn= 2.281E+06
28 2.28138e+06 6.14875e+01
Cost_fcn eval took 6.88 sec - Cost_fcn= 2.173E+06
29 2.17303e+06 6.00497e+01
Cost_fcn eval took 7.09 sec - Cost_fcn= 2.152E+06
30 2.15206e+06 5.96543e+01
Cost_fcn eval took 6.88 sec - Cost_fcn= 2.142E+06
31 2.14197e+06 5.95280e+01
Cost_fcn eval took 6.47 sec - Cost_fcn= 2.118E+06
32 2.11807e+06 5.90464e+01
Cost_fcn eval took 7.01 sec - Cost_fcn= 2.114E+06
33 2.11441e+06 5.90031e+01
Cost_fcn eval took 6.86 sec - Cost_fcn= 2.102E+06
34 2.10188e+06 5.87821e+01
Cost_fcn eval took 6.77 sec - Cost_fcn= 2.024E+06
35 2.02366e+06 5.76671e+01
Iterative inversion took :- 00:04:56
END OF TEST
minimum value = 2023663.4048203537
result code = 5There you have it ✌️
This repository uses data from the SEG Open Data collection, specifically the Elastic Marmousi model. The data has been resampled for use in this project. The original data is provided by the Allied Geophysical Laboratory of the University of Houston and it is licensed under the Creative Commons Attribution 4.0 International License. You can download the original data from the following link:
Data License
The resampled data used in this repository is derived from the Elastic Marmousi model data, which is licensed under the Creative Commons Attribution 4.0 International License. For more information about the license, please visit: Creative Commons Attribution 4.0 International License
Code License
The code in this repository is licensed under the Apache License 2.0. You are free to use, modify, and distribute the code under the terms of this license. For the full text of the license, please visit: Apache License 2.0