fourndo/simpeg_scripts

Repository of scripts to run SimPEG inversions and forwards

simpeg_scripts's Issues

add bounds option

in the json file
"bounds":[-1.5, 0.2] #lower, upper

for gravity something like this works, add after line 954 in 20191113 script

if "bounds" in list(input_dict.keys()):    
    if input_dict["inversion_type"] == 'grav':
        lower_grav_bound = input_dict["bounds"][0]
        upper_grav_bound = input_dict["bounds"][1]
        opt = Optimization.ProjectedGNCG(maxIter=max_global_iterations, lower=lower_grav_bound,
                                         upper=upper_grav_bound, maxIterLS=20,
                                         maxIterCG=30, tolCG=1e-3)
    else:
        opt = Optimization.ProjectedGNCG(maxIter=max_global_iterations, lower=-np.inf,
                                         upper=np.inf, maxIterLS=20,
                                         maxIterCG=30, tolCG=1e-3)
else:
    opt = Optimization.ProjectedGNCG(maxIter=max_global_iterations, lower=-np.inf,
                                     upper=np.inf, maxIterLS=20,
                                     maxIterCG=30, tolCG=1e-3)

then extend to mag, mvi, mvi_s

alphas on MVI-S

Are we missing setting the alphas for amplitude on the MVI-S:

    reg_a = Regularization.Sparse(mesh, indActive=activeCells,
                                  mapping=wires.amp, gradientType=gradient_type)
    reg_a.norms = model_norms[:4].T
    reg_a.mref = mref

Should we include reg_a.alphas = alphas?

parallelized on gravity inversion : keyword in Json file

gravity parallelization requires keyword 'dask', rather than a boolean 'True'. see gravity.py line 27

Working directory

How is the working directory supposed to work? It seems to try to pull from the input file but reverts to the root for me '###-\MVI_C_20m\mvi.mod' in Spyder.

This fixes it for me:

    if len(workDir) > 0:
        workDir += dsep
    else:
        workDir = os.getcwd() + dsep
    print('Working directory:', workDir)

but maybe I am just using it wrong?

reg.mref

I see we still have this issue:

    reg_t.mref = mref

    # Assemble the 3-component regularizations
    reg = reg_p + reg_s + reg_t
    reg.mref = mref

Does reg.mref = mref overwrite reg_t.mref? Are they both used?