Note
Click here to download the full example code
1.2: Data Structure and Manipulation¶
# Importing GemPy
import gempy as gp
import gempy
# Importing auxiliary libraries
import numpy as np
import pandas as pd
pd.set_option('precision', 2)
Series¶
Series is the object that contains the properties associated with each independent scalar field. Right now it is simply the order of the series (which is inferred by the index order). But in the future will be add the unconformity relation or perhaps the type of interpolator
Series and Faults classes are quite entangled since fault are a view of series
We can modify the series bt using set_series_index:
series.set_series_index(['foo', 'foo2', 'foo5', 'foo7'])
series
The index of series are pandas categories. These provides quite handy backend functionality (see pandas.Categorical).
Out:
CategoricalIndex(['foo', 'foo2', 'foo5', 'foo7'], categories=['foo', 'foo2', 'foo5', 'foo7'], ordered=False, dtype='category')
For adding new series:
series.add_series('foo3')
series
Delete series
series.delete_series('foo3')
series
Rename series:
series.rename_series({'foo': 'boo'})
series
Reorder series:
series.reorder_series(['foo2', 'boo', 'foo7', 'foo5'])
series
Faults¶
The df faults is used to characterize which mathematical series behave as fault and if mentioned faults are finite or infinite. Both dataframes get updated automatically as we modify the series object linked to the fault object (by passing it when a Series object is created).
Finally we have the faults relations df which captures which mathematical series a given fault offset in order to reproduce complex faulting networks
We can use set_is_fault to choose which of our series are faults:
faults.set_is_fault(['boo'])
Similar thing for the fault relations:
Now if we change the series df and we update the series already defined will conserve their values while the new ones will be set to false:
series.add_series('foo20')
When we add new series the values switch to NaN. We will be careful not having any NaNs in the DataFrames or we will raise errors down the line.
Surfaces:¶
The df surfaces contains three properties. id refers to the order of the surfaces on the sequential pile, i.e. the strict order of computation. values on the other hand is the final value that each voxel will have after discretization. This may be useful for example in the case we want to map a specific geophysical property (such as density) to a given unit. By default both are the same since to discretize lithological units the value is arbitrary.
From an empty df¶
The Surfaces class needs to have an associate series object. This will
limit the name of the series since they are a pandas.Categorical.
We can set any number of formations by passing a list with the names. By default they will take the name or the first series.
surfaces.set_surfaces_names(['foo', 'foo2', 'foo5'])
We can add new formations:
surfaces.add_surface(['feeeee'])
surfaces
The column formation is also a ‘’pandas.Categorical’’. This will be important for the Data classes (surface_points and Orientations)
surfaces.df['surface']
Out:
0 foo
1 foo2
2 foo5
3 feeeee
Name: surface, dtype: object
Set values¶
To set the values we do it with the following method
surfaces.set_surfaces_values([2, 2, 2, 5])
Set values with a given name:¶
We can give specific names to the properties (i.e. density)
surfaces.add_surfaces_values([[2, 2, 2, 6], [2, 2, 1, 8]], ['val_foo', 'val2_foo'])
Delete formations values¶
To delete a full property:
surfaces.delete_surface_values(['val_foo', 'value_0'])
Out:
True
Set formation values¶
We can also use set_surface_values instead adding. This will delete the previous
properties and add the new one
surfaces.set_surfaces_values([[2, 2, 2, 6], [2, 2, 1, 8]], ['val_foo', 'val2_foo'])
surfaces
The last property is the correspondant series that each formation belong
to. series and formation are pandas categories. To get a
overview of what this mean check
https://pandas.pydata.org/pandas-docs/stable/categorical.html.
surfaces.df['series']
Out:
0 foo20
1 foo20
2 foo20
3 foo20
Name: series, dtype: category
Categories (5, object): ['foo2', 'boo', 'foo7', 'foo5', 'foo20']
surfaces.df['surface']
Out:
0 foo
1 foo2
2 foo5
3 feeeee
Name: surface, dtype: object
Map series to formation¶
To map a series to a formation we can do it by passing a dict:
If a series does not exist in the Series object, we rise a warning
and we set those formations to nans
d = {"foo7": 'foo', "booX": ('foo2', 'foo5', 'fee')}
surfaces.map_series({"foo7": 'foo', "boo": ('foo2', 'foo5', 'fee')})
An advantage of categories is that they are order so no we can tidy the df by series and formation
Modify surface color¶
The surfaces DataFrame also contains a column for the color in which the surfaces are displayed. To change the color, call
surfaces.colors.change_colors()
Out:
Click to select new colors.
VBox(children=(ColorPicker(value='#015482', description='foo'), ColorPicker(value='#9f0052', description='foo2'), ColorPicker(value='#ffbe00', description='foo5'), ColorPicker(value='#728f02', description='feeeee')))
This allow to change the colors interactively. If you already know which colors you want to use, you can also update them with a dictionary mapping the surface name to a hex color string:
new_colors = {'foo': '#ff8000', 'foo5': '#4741be'}
surfaces.colors.change_colors(new_colors)
Data¶
surface_points¶
These two DataFrames (df from now on) will contain the individual information of each point at an interface or orientation. Some properties of this table are mapped from the df below.
surface_points.set_surface_points(pd.DataFrame(np.random.rand(6, 3)),
['foo', 'foo5', 'lala', 'foo5', 'lala', 'feeeee'])
surface_points.map_data_from_surfaces(surfaces, 'series')
surface_points
surface_points.map_data_from_surfaces(surfaces, 'id')
surface_points
surface_points.map_data_from_series(series, 'order_series')
surface_points
surface_points.sort_table()
surface_points
Set values passing pole vectors:¶
orientations.set_orientations(np.random.rand(6, 3) * 10,
np.random.rand(6, 3),
surface=['foo', 'foo5', 'lala', 'foo5',
'lala', 'feeeee'])
Set values pasing orientation data: azimuth, dip, pole (dip direction)¶
orientations.set_orientations(np.random.rand(6, 3) * 10,
orientation=np.random.rand(6, 3) * 20,
surface=['foo', 'foo5', 'lala', 'foo5',
'lala', 'feeeee'])
Mapping data from the other df¶
orientations.map_data_from_surfaces(surfaces, 'series')
orientations
orientations.map_data_from_surfaces(surfaces, 'id')
orientations
orientations.map_data_from_series(series, 'order_series')
orientations
Grid¶
grid = gp.Grid()
grid.create_regular_grid([0, 10, 0, 10, 0, 10], [50, 50, 50])
Out:
<gempy.core.grid_modules.grid_types.RegularGrid object at 0x7f323c6e9a90>
Out:
array([[0.1, 0.1, 0.1],
[0.1, 0.1, 0.3],
[0.1, 0.1, 0.5],
...,
[9.9, 9.9, 9.5],
[9.9, 9.9, 9.7],
[9.9, 9.9, 9.9]])
Rescaling Data¶
rescaling = gempy.core.data_modules.geometric_data.ScalingSystem(
surface_points, orientations, grid)
Additional Data¶
ad = gp.AdditionalData(surface_points, orientations, grid, faults, surfaces, rescaling)
Out:
dtype category
output category
theano_optimizer category
device category
verbosity object
dtype: object
ad.rescaling_data
Interpolator¶
faults.df['isFault'].values
Out:
array([False, True, False, False, False])
Out:
Compiling theano function...
Level of Optimization: fast_compile
Device: cpu
Precision: float64
Number of faults: 1
Compilation Done!
<theano.compile.function_module.Function object at 0x7f321aa5cbb0>
Out:
len sereies i [0 2]
len sereies o [0 4]
len sereies w [ 0 17]
n surfaces per series [0 2]
n universal eq [3]
is finite [0 0 0 0 0]
is erosion [0]
is onlap [0]
Total running time of the script: ( 0 minutes 3.277 seconds)
