Using the tdManager to investigate inversion results

One of the purposes of the tdManager is to provide an easy interface to analyze inversion results. Inversion results can either be directly generated using the crtomo.tdMan.invert() function, or by loading from an already-finished inversion directory (the tomodir).

In this example we will explore the various functionalities of the tdManager.

we need the crtomo module to continue

import crtomo
# the pprint module is used just to prettify the output of this example, please
# ignore for further use
import pprint

load inversion results from an already-finished tomodir

tdm = crtomo.tdMan(tomodir='tomodir')

importing tomodir tomodir
This grid was sorted using CutMcK. The nodes were resorted!
Triangular grid found
importing tomodir results
loaded configs: (220, 2)
reading voltages
Reading inversion results
is robust True

The first order of business is to access the convergence behavior of the inversion. This data is loaded into a pandas.DataFrame:

print(tdm.inv_stats)

# TODO Plot inv stats

    iteration  main_iteration it_type  ... steplength  stepsize  l1ratio
0           0               0   DC/IP  ...        NaN       NaN     1.14
1           1               1   DC/IP  ...        NaN  13200.00      NaN
2           2               1   DC/IP  ...        NaN  13600.00      NaN
3           3               1   DC/IP  ...        NaN  14200.00      NaN
4           4               1   DC/IP  ...        NaN  15300.00      NaN
..        ...             ...     ...  ...        ...       ...      ...
75          8               8   DC/IP  ...        NaN     48.90      NaN
76          9               8   DC/IP  ...        NaN     52.60      NaN
77         10               8   DC/IP  ...        NaN     56.60      NaN
78         11               8   DC/IP  ...        NaN     26.30      NaN
79          8               8   DC/IP  ...        0.0     52.63     1.00

[80 rows x 14 columns]

Inversion results are stored as parameter sets in the ParMan instance (crtomo.ParMan):

print(tdm.parman)

<crtomo.parManager.ParMan object at 0x7f6761c941c0>

the ParMan stores multiple parameter sets (i.e., data sets which assign each cell of the FE-mesh a given value), identified by an integer index number, the parameter id, pid. The pids for the various parameters imported from the inversion results are stored in dictionary entries:

pprint.pprint(tdm.a['inversion'])

# We can see several entries in the dictionary: The resistivity magnitude
# (rmag), the phase value (rpha), the real part of the complex conductivity
# (cre), the imaginary part of the complex conductivity (cim) and an entry
# which combines cre and cim (cre_cim). For each inversion, the parameter sets
# have an own pid to make the results accessible.

{'cim': [19, 21, 23, 25, 27, 29, 31, 33, 35],
 'cre': [18, 20, 22, 24, 26, 28, 30, 32, 34],
 'cre_cim': [[18, 19],
             [20, 21],
             [22, 23],
             [24, 25],
             [26, 27],
             [28, 29],
             [30, 31],
             [32, 33],
             [34, 35]],
 'fwd_response_rmag': [2, 5, 8, 11, 14, 17, 20, 23, 26],
 'fwd_response_rpha': [3, 6, 9, 12, 15, 18, 21, 24, 27],
 'fwd_response_wdfak': [4, 7, 10, 13, 16, 19, 22, 25, 28],
 'l1_dw_log10_norm': 37,
 'l2_dw_log10_norm': 38,
 'resm': 36,
 'rmag': [0, 1, 2, 3, 4, 5, 6, 7, 8],
 'rpha': [9, 10, 11, 12, 13, 14, 15, 16, 17]}

To access a given parameter set, extract the pid from the dict and then access the data in the ParMan instance. If we would like to access the phase values for e.g. the second inversion run, we have to choose the index accordingly:

pid_rpha_2nd = tdm.a['inversion']['rpha'][1]
print('pid of the second phase inversion result: ', pid_rpha_2nd)

rpha_second_inversion = tdm.parman.parsets[pid_rpha_2nd]
print(
    'Pixel values of the second phase inverion result: ',
    rpha_second_inversion
)

pid of the second phase inversion result:  10
Pixel values of the second phase inverion result:  [ -7.50958061 -14.9736643  -10.9377241  ...  -2.00979733  -2.66891265
 -12.5039339 ]

The last inversion result can be accessed using the index -1

rpha_last_inversion = tdm.parman.parsets[tdm.a['inversion']['rmag'][-1]]

Most of the time, it is of interest to get the final inversion result. This can be done in two seperate ways. Either by settings the index to [-1] or by using the built-in retriever function from crtomo-tools:

rmag_final_inversion = tdm.inv_last_rmag_parset()

# The function returns a numpy-array with the inversion results. A similar
# function exists to extract the final results for the phase value (rpha),
# the real part of the complex conductivity (cre) and the imaginary
# part of the complex conductivity (cim):

rpha_final_inversion = tdm.inv_last_rpha_parset()
cre_final_inversion = tdm.inv_last_cre_parset()
cim_final_inversion = tdm.inv_last_cim_parset()

TODO: Extract from point, line, polygon, rectangle