from collections import Iterable
from typing import Union
from gempy.core.data import Grid, Surfaces, AdditionalData
from gempy.core.data_modules.geometric_data import SurfacePoints, Orientations
from gempy.core.data_modules.stack import Faults, Series
from gempy.utils.meta import _setdoc_pro, _setdoc
import gempy.utils.docstring as ds
import numpy as np
import theano
@_setdoc_pro(
[SurfacePoints.__doc__, Orientations.__doc__, Grid.__doc__, Surfaces.__doc__, Series.__doc__,
Faults.__doc__, AdditionalData.__doc__])
class Interpolator(object):
"""Class that act as:
1) linker between the data objects and the theano graph
2) container of theano graphs + shared variables
3) container of theano function
Args:
surface_points (SurfacePoints): [s0]
orientations (Orientations): [s1]
grid (Grid): [s2]
surfaces (Surfaces): [s3]
series (Series): [s4]
faults (Faults): [s5]
additional_data (AdditionalData): [s6]
kwargs:
- compile_theano: if true, the function is compile at the creation of the class
Attributes:
surface_points (SurfacePoints)
orientations (Orientations)
grid (Grid)
surfaces (Surfaces)
faults (Faults)
additional_data (AdditionalData)
dtype (['float32', 'float64']): float precision
theano_graph: theano graph object with the properties from AdditionalData -> Options
theano function: python function to call the theano code
"""
# TODO assert passed data is rescaled
def __init__(self, surface_points: "SurfacePoints", orientations: "Orientations", grid: "Grid",
surfaces: "Surfaces", series: Series, faults: "Faults",
additional_data: "AdditionalData", **kwargs):
# Test
self.surface_points = surface_points
self.orientations = orientations
self.grid = grid
self.additional_data = additional_data
self.surfaces = surfaces
self.series = series
self.faults = faults
self.dtype = additional_data.options.df.loc['values', 'dtype']
self.theano_graph = self.create_theano_graph(additional_data, inplace=False)
self.theano_function = None
self._compute_len_series()
def _compute_len_series(self):
self.len_series_i = self.additional_data.structure_data.df.loc[
'values', 'len series surface_points'] - \
self.additional_data.structure_data.df.loc[
'values', 'number surfaces per series']
if self.len_series_i.shape[0] == 0:
self.len_series_i = np.zeros(1, dtype=int)
self.len_series_o = self.additional_data.structure_data.df.loc[
'values', 'len series orientations'].astype(
'int32')
if self.len_series_o.shape[0] == 0:
self.len_series_o = np.zeros(1, dtype=int)
self.len_series_u = self.additional_data.kriging_data.df.loc[
'values', 'drift equations'].astype('int32')
if self.len_series_u.shape[0] == 0:
self.len_series_u = np.zeros(1, dtype=int)
self.len_series_f = self.faults.faults_relations_df.sum(axis=0).values.astype('int32')[
:self.additional_data.get_additional_data()['values'][
'Structure', 'number series']]
if self.len_series_f.shape[0] == 0:
self.len_series_f = np.zeros(1, dtype=int)
self.len_series_w = self.len_series_i + self.len_series_o * 3 + self.len_series_u + self.len_series_f
@_setdoc_pro([AdditionalData.__doc__, ds.inplace, ds.theano_graph_pro])
def create_theano_graph(self, additional_data: "AdditionalData" = None, inplace=True,
output=None, **kwargs):
"""
Create the graph accordingly to the options in the AdditionalData object
Args:
additional_data (AdditionalData): [s0]
inplace (bool): [s1]
Returns:
TheanoGraphPro: [s2]
"""
if output is None:
output = ['geology']
import gempy.core.theano_modules.theano_graph_pro as tg
import importlib
importlib.reload(tg)
if additional_data is None:
additional_data = self.additional_data
self.dtype = additional_data.options.df.loc['values', 'dtype']
graph = tg.TheanoGraphPro(
optimizer=additional_data.options.df.loc['values', 'theano_optimizer'],
verbose=additional_data.options.df.loc['values', 'verbosity'],
output=output,
**kwargs)
if inplace is True:
self.theano_graph = graph
else:
return graph
@_setdoc_pro([ds.theano_graph_pro])
def set_theano_graph(self, th_graph):
"""
Attach an already create theano graph.
Args:
th_graph (TheanoGraphPro): [s0]
Returns:
True
"""
self.theano_graph = th_graph
return True
def set_theano_shared_kriging(self):
"""
Set to the theano_graph attribute the shared variables of kriging values from the linked
:class:`AdditionalData`.
Returns:
True
"""
# Range
range_val = self.additional_data.kriging_data.df.loc['values', 'range']
range_res = range_val / self.additional_data.rescaling_data.df.loc[
'values', 'rescaling factor']
range_list = range_res * np.ones(
self.additional_data.structure_data.df.loc['values',
'number series'])
#
# if type(range_res) is np.float or type(range_res) is np.float64:
# range_list = range_res * np.ones(
# self.additional_data.structure_data.df.loc['values',
# 'number series']
# )
# elif isinstance(range_res, Iterable):
# range_list = range_res
# else:
# raise AttributeError('Range must be either int or Iterable')
# TODO add rescaled range and co into the rescaling data df?
self.theano_graph.a_T.set_value(np.cast[self.dtype](range_list))
# Covariance at 0
cov_val = self.additional_data.kriging_data.df.loc['values', '$C_o$']
cov_res = cov_val / self.additional_data.rescaling_data.df.loc[
'values', 'rescaling factor']
cov_list = cov_res * np.ones(
self.additional_data.structure_data.df.loc['values',
'number series']
)
#
#
# if type(cov_res) is np.float or type(cov_res) is np.float64:
# cov_list = cov_res * np.ones(
# self.additional_data.structure_data.df.loc['values',
# 'number series']
# )
# elif isinstance(cov_res, Iterable):
# cov_list = cov_res
# else:
# raise AttributeError('Covariance at 0 must be either int or Iterable')
self.theano_graph.c_o_T.set_value(
np.cast[self.dtype](cov_list)
)
# universal grades
self.theano_graph.n_universal_eq_T.set_value(
list(self.additional_data.kriging_data.df.loc['values', 'drift equations'].astype(
'int32')[self.non_zero]))
self.set_theano_shared_nuggets()
def set_theano_shared_nuggets(self):
# nugget effect
# len_orientations = self.additional_data.structure_data.df.loc['values', 'len series orientations']
# len_orientations_len = np.sum(len_orientations)
self.theano_graph.nugget_effect_grad_T.set_value(
np.cast[self.dtype](np.tile(
self.orientations.df['smooth'], 3)))
# len_rest_form = (self.additional_data.structure_data.df.loc['values', 'len surfaces surface_points'])
# len_rest_len = np.sum(len_rest_form)
self.theano_graph.nugget_effect_scalar_T.set_value(
np.cast[self.dtype](self.surface_points.df['smooth']))
return True
def set_theano_shared_structure_surfaces(self):
"""
Set to the theano_graph attribute the shared variables of structure from the linked
:class:`AdditionalData`.
Returns:
True
"""
len_rest_form = (self.additional_data.structure_data.df.loc[
'values', 'len surfaces surface_points'] - 1)
self.theano_graph.number_of_points_per_surface_T.set_value(len_rest_form.astype('int32'))
class InterpolatorWeights(Interpolator):
def __init__(self, surface_points: "SurfacePoints", orientations: "Orientations", grid: "Grid",
surfaces: "Surfaces", series, faults: "Faults", additional_data: "AdditionalData",
**kwargs):
super(InterpolatorWeights, self).__init__(surface_points, orientations, grid, surfaces,
series, faults,
additional_data, **kwargs)
def get_python_input_weights(self, fault_drift=None):
"""
Get values from the data objects used during the interpolation:
- dip positions XYZ
- dip angles
- azimuth
- polarity
- surface_points coordinates XYZ
Returns:
(list)
"""
# orientations, this ones I tile them inside theano. PYTHON VAR
dips_position = self.orientations.df[['X_c', 'Y_c', 'Z_c']].values
dip_angles = self.orientations.df["dip"].values
azimuth = self.orientations.df["azimuth"].values
polarity = self.orientations.df["polarity"].values
surface_points_coord = self.surface_points.df[['X_c', 'Y_c', 'Z_c']].values
if fault_drift is None:
fault_drift = np.zeros((0, self.grid.values.shape[0] + 2 * self.len_series_i.sum()))
# fault_drift = np.zeros((0, surface_points_coord.shape[0]))
# Set all in a list casting them in the chosen dtype
idl = [np.cast[self.dtype](xs) for xs in
(dips_position, dip_angles, azimuth, polarity, surface_points_coord,
fault_drift)]
return idl
def compile_th_fn(self, inplace=False, debug=False):
self.set_theano_shared_kriging()
self.set_theano_shared_structure_surfaces()
# This are the shared parameters and the compilation of the function. This will be hidden as well at some point
input_data_T = self.theano_graph.input_parameters_kriging
print('Compiling theano function...')
th_fn = theano.function(input_data_T,
self.theano_graph.compute_weights(),
# mode=NanGuardMode(nan_is_error=True),
on_unused_input='warn',
allow_input_downcast=False,
profile=False)
if inplace is True:
self.theano_function = th_fn
if debug is True:
print('Level of Optimization: ', theano.config.optimizer)
print('Device: ', theano.config.device)
print('Precision: ', self.dtype)
print('Number of faults: ',
self.additional_data.structure_data.df.loc['values', 'number faults'])
print('Compilation Done!')
return th_fn
class InterpolatorScalar(Interpolator):
def __init__(self, surface_points: "SurfacePoints", orientations: "Orientations", grid: "Grid",
surfaces: "Surfaces", series, faults: "Faults", additional_data: "AdditionalData",
**kwargs):
super(InterpolatorScalar, self).__init__(surface_points, orientations, grid, surfaces,
series, faults,
additional_data, **kwargs)
def get_python_input_zx(self, fault_drift=None):
"""
Get values from the data objects used during the interpolation:
- dip positions XYZ
- dip angles
- azimuth
- polarity
- surface_points coordinates XYZ
Returns:
(list)
"""
# orientations, this ones I tile them inside theano. PYTHON VAR
dips_position = self.orientations.df[['X_c', 'Y_c', 'Z_c']].values
dip_angles = self.orientations.df["dip"].values
azimuth = self.orientations.df["azimuth"].values
polarity = self.orientations.df["polarity"].values
surface_points_coord = self.surface_points.df[['X_c', 'Y_c', 'Z_c']].values
grid = self.grid.values_c
if fault_drift is None:
fault_drift = np.zeros((0, grid.shape[0] + 2 * self.len_series_i.sum()))
# fault_drift = np.zeros((0, grid.shape[0] + surface_points_coord.shape[0]))
# Set all in a list casting them in the chosen dtype
idl = [np.cast[self.dtype](xs) for xs in
(dips_position, dip_angles, azimuth, polarity, surface_points_coord,
fault_drift, grid)]
return idl
def compile_th_fn(self, weights=None, grid=None, inplace=False, debug=False):
"""
Args:
weights: Constant weights
grid: Constant grids
inplace:
debug:
Returns:
"""
self.set_theano_shared_kriging()
self.set_theano_shared_structure_surfaces()
# This are the shared parameters and the compilation of the function. This will be hidden as well at some point
input_data_T = self.theano_graph.input_parameters_kriging_export
print('Compiling theano function...')
if weights is None:
weights = self.theano_graph.compute_weights()
else:
weights = theano.shared(weights)
if grid is None:
grid = self.theano_graph.grid_val_T
else:
grid = theano.shared(grid)
th_fn = theano.function(input_data_T,
self.theano_graph.compute_scalar_field(weights, grid),
# mode=NanGuardMode(nan_is_error=True),
on_unused_input='ignore',
allow_input_downcast=False,
profile=False)
if inplace is True:
self.theano_function = th_fn
if debug is True:
print('Level of Optimization: ', theano.config.optimizer)
print('Device: ', theano.config.device)
print('Precision: ', theano.config.floatX)
print('Number of faults: ',
self.additional_data.structure_data.df.loc['values', 'number faults'])
print('Compilation Done!')
return th_fn
class InterpolatorBlock(Interpolator):
def __init__(self, surface_points: "SurfacePoints", orientations: "Orientations", grid: "Grid",
surfaces: "Surfaces", series: Series, faults: "Faults",
additional_data: "AdditionalData", **kwargs):
super(InterpolatorBlock, self).__init__(surface_points, orientations, grid, surfaces,
series,
faults, additional_data, **kwargs)
self.theano_function_formation = None
self.theano_function_faults = None
def get_python_input_block(self, fault_drift=None):
"""
Get values from the data objects used during the interpolation:
- dip positions XYZ
- dip angles
- azimuth
- polarity
- surface_points coordinates XYZ
Returns:
(list)
"""
# orientations, this ones I tile them inside theano. PYTHON VAR
dips_position = self.orientations.df[['X_c', 'Y_c', 'Z_c']].values
dip_angles = self.orientations.df["dip"].values
azimuth = self.orientations.df["azimuth"].values
polarity = self.orientations.df["polarity"].values
surface_points_coord = self.surface_points.df[['X_c', 'Y_c', 'Z_c']].values
grid = self.grid.values_c
if fault_drift is None:
fault_drift = np.zeros((0, grid.shape[0] + 2 * self.len_series_i.sum()))
values_properties = self.surfaces.df.iloc[:, self.surfaces._n_properties:].values.astype(
self.dtype).T
# Set all in a list casting them in the chosen dtype
idl = [np.cast[self.dtype](xs) for xs in
(dips_position, dip_angles, azimuth, polarity, surface_points_coord,
fault_drift, grid, values_properties)]
return idl
def compile_th_fn_formation_block(self, Z_x=None, weights=None, grid=None,
values_properties=None, inplace=False,
debug=False):
"""
Args:
weights: Constant weights
grid: Constant grids
inplace:
debug:
Returns:
"""
self.set_theano_shared_kriging()
self.set_theano_shared_structure_surfaces()
# This are the shared parameters and the compilation of the function. This will be hidden as well at some point
input_data_T = self.theano_graph.input_parameters_block
print('Compiling theano function...')
if weights is None:
weights = self.theano_graph.compute_weights()
else:
weights = theano.shared(weights)
if grid is None:
grid = self.theano_graph.grid_val_T
else:
grid = theano.shared(grid)
if values_properties is None:
values_properties = self.theano_graph.values_properties_op
else:
values_properties = theano.shared(values_properties)
if Z_x is None:
Z_x = self.theano_graph.compute_scalar_field(weights, grid)
else:
Z_x = theano.shared(Z_x)
th_fn = theano.function(input_data_T,
self.theano_graph.compute_formation_block(
Z_x,
self.theano_graph.get_scalar_field_at_surface_points(Z_x),
values_properties
),
on_unused_input='ignore',
allow_input_downcast=False,
profile=False)
if inplace is True:
self.theano_function_formation = th_fn
if debug is True:
print('Level of Optimization: ', theano.config.optimizer)
print('Device: ', theano.config.device)
print('Precision: ', self.dtype)
print('Number of faults: ',
self.additional_data.structure_data.df.loc['values', 'number faults'])
print('Compilation Done!')
return th_fn
def compile_th_fn_fault_block(self, Z_x=None, weights=None, grid=None, values_properties=None,
inplace=False, debug=False):
"""
Args:
weights: Constant weights
grid: Constant grids
inplace:
debug:
Returns:
"""
self.set_theano_shared_kriging()
self.set_theano_shared_structure_surfaces()
# This are the shared parameters and the compilation of the function. This will be hidden as well at some point
input_data_T = self.theano_graph.input_parameters_block
print('Compiling theano function...')
if weights is None:
weights = self.theano_graph.compute_weights()
else:
weights = theano.shared(weights)
if grid is None:
grid = self.theano_graph.grid_val_T
else:
grid = theano.shared(grid)
if values_properties is None:
values_properties = self.theano_graph.values_properties_op
else:
values_properties = theano.shared(values_properties)
if Z_x is None:
Z_x = self.theano_graph.compute_scalar_field(weights, grid)
else:
Z_x = theano.shared(Z_x)
th_fn = theano.function(input_data_T,
self.theano_graph.compute_fault_block(
Z_x,
self.theano_graph.get_scalar_field_at_surface_points(Z_x),
values_properties,
0,
grid
),
# mode=NanGuardMode(nan_is_error=True),
on_unused_input='ignore',
allow_input_downcast=False,
profile=False)
if inplace is True:
self.theano_function_faults = th_fn
if debug is True:
print('Level of Optimization: ', theano.config.optimizer)
print('Device: ', theano.config.device)
print('Precision: ', self.dtype)
print('Number of faults: ',
self.additional_data.structure_data.df.loc['values', 'number faults'])
print('Compilation Done!')
return th_fn
[docs]class InterpolatorGravity:
[docs] def set_theano_shared_tz_kernel(self, tz=None):
"""Set the theano component tz to each voxel"""
if tz is None or tz == 'auto':
try:
tz = self.calculate_tz(self.grid.centered_grid)
except AttributeError:
raise AttributeError('You need to calculate or pass tz first.')
self.theano_graph.tz.set_value(tz.astype(self.dtype))
def calculate_tz(self, centered_grid):
from gempy.assets.geophysics import GravityPreprocessing
g = GravityPreprocessing(centered_grid)
return g.set_tz_kernel()
def set_theano_shared_pos_density(self, pos_density):
self.theano_graph.pos_density.set_value(pos_density)
def set_theano_shared_l0_l1(self):
self.theano_graph.lg0.set_value(self.grid.get_grid_args('centered')[0])
self.theano_graph.lg1.set_value(self.grid.get_grid_args('centered')[1])
def set_theano_shared_gravity(self, tz='auto', pos_density=1):
self.set_theano_shared_tz_kernel(tz)
self.set_theano_shared_pos_density(pos_density)
self.set_theano_shared_l0_l1()
class InterpolatorMagnetics:
def set_theano_shared_Vs_kernel(self, V=None):
if V is None or V == 'auto':
try:
V = self.calculate_V(self.grid.centered_grid)
except AttributeError:
raise AttributeError('You need to calculate or pass V first.')
self.theano_graph.V.set_value(V.astype(self.dtype))
def calculate_V(self, centered_grid):
from gempy.assets.geophysics import MagneticsPreprocessing
Vmodel = MagneticsPreprocessing(centered_grid).set_Vs_kernel()
return Vmodel
def set_theano_shared_pos_magnetics(self, pos_magnetics):
self.theano_graph.pos_magnetics.set_value(pos_magnetics)
def set_theano_shared_magnetic_cts(self, incl, decl, B_ext=52819.8506939139e-9):
"""
Args:
B_ext : External magnetic field in [T], in magnetic surveys this is the geomagnetic field - varies temporaly
incl : Dip of the geomagnetic field in degrees- varies spatially
decl : Angle between magnetic and true North in degrees - varies spatially
"""
self.theano_graph.incl.set_value(incl)
self.theano_graph.decl.set_value(decl)
self.theano_graph.B_ext.set_value(B_ext)
def set_theano_shared_l0_l1(self):
self.theano_graph.lg0.set_value(self.grid.get_grid_args('centered')[0])
self.theano_graph.lg1.set_value(self.grid.get_grid_args('centered')[1])
def set_theano_shared_magnetics(self, V='auto', pos_magnetics=1,
incl=None, decl=None, B_ext=52819.8506939139e-9):
self.set_theano_shared_Vs_kernel(V)
self.set_theano_shared_pos_magnetics(pos_magnetics)
self.set_theano_shared_magnetic_cts(incl, decl, B_ext)
self.set_theano_shared_l0_l1()
[docs]@_setdoc_pro(ds.ctrl)
@_setdoc([Interpolator.__doc__])
class InterpolatorModel(Interpolator, InterpolatorGravity, InterpolatorMagnetics):
"""
Child class of :class:`Interpolator` which set the shared variables and compiles the theano
graph to compute the geological model, i.e. lithologies.
Attributes:
compute_weights_ctrl (list[bool]): [s0]
compute_scalar_ctrl (list[bool]):
compute_block_ctrl (list[bool]):
Interpolator Doc
"""
[docs] def __init__(self, surface_points: "SurfacePoints", orientations: "Orientations", grid: "Grid",
surfaces: "Surfaces", series, faults: "Faults", additional_data: "AdditionalData",
**kwargs):
super().__init__(surface_points, orientations, grid, surfaces, series, faults,
additional_data, **kwargs)
self.len_series_i = np.zeros(1)
self.len_series_o = np.zeros(1)
self.len_series_u = np.zeros(1)
self.len_series_f = np.zeros(1)
self.len_series_w = np.zeros(1)
self.set_initial_results()
n_series = 1000
self.compute_weights_ctrl = np.ones(n_series, dtype=bool)
self.compute_scalar_ctrl = np.ones(n_series, dtype=bool)
self.compute_block_ctrl = np.ones(n_series, dtype=bool)
[docs] def reset_flow_control_initial_results(self, reset_weights=True, reset_scalar=True,
reset_block=True):
"""
Method to reset to the initial state all the recompute ctrl. After calling this method next time
gp.compute_model is called, everything will be computed. Panic bottom.
Args:
reset_weights (bool):
reset_scalar (bool):
reset_block (bool):
Returns:
True
"""
n_series = self.len_series_i.shape[0]
x_to_interp_shape = self.grid.values_c.shape[0] + 2 * self.len_series_i.sum()
if reset_weights is True:
self.compute_weights_ctrl = np.ones(1000, dtype=bool)
self.theano_graph.weights_vector.set_value(
np.zeros((self.len_series_w.sum()), dtype=self.dtype))
if reset_scalar is True:
self.compute_scalar_ctrl = np.ones(1000, dtype=bool)
self.theano_graph.scalar_fields_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype=self.dtype))
if reset_block is True:
self.compute_block_ctrl = np.ones(1000, dtype=bool)
self.theano_graph.mask_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype='bool'))
self.theano_graph.block_matrix.set_value(
np.zeros((n_series,
self.surfaces.df.iloc[:, self.surfaces._n_properties:].values.shape[1],
x_to_interp_shape), dtype=self.dtype))
return True
[docs] def set_flow_control(self):
"""
Initialize the ctrl vectors to the number of series size.
Returns:
True
"""
n_series = 1000
self.compute_weights_ctrl = np.ones(n_series, dtype=bool)
self.compute_scalar_ctrl = np.ones(n_series, dtype=bool)
self.compute_block_ctrl = np.ones(n_series, dtype=bool)
return True
[docs] @_setdoc_pro(reset_flow_control_initial_results.__doc__)
def set_all_shared_parameters(self, reset_ctrl=False):
"""
Set all theano shared parameters required for the computation of lithology
Args:
reset_ctrl (bool): If true, [s0]
Returns:
True
"""
self.set_theano_shared_loop()
self.set_theano_shared_relations()
self.set_theano_shared_kriging()
self.set_theano_shared_structure_surfaces()
# self.set_theano_shared_topology()
if reset_ctrl is True:
self.reset_flow_control_initial_results()
return True
def set_theano_shared_topology(self):
max_lith = self.surfaces.df.groupby('isFault')['id'].count()[False]
if type(max_lith) != int:
max_lith = 0
self.theano_graph.max_lith.set_value(max_lith)
self.theano_graph.regular_grid_res.set_value(self.grid.regular_grid.resolution)
self.theano_graph.dxdydz.set_value(
np.array(self.grid.regular_grid.get_dx_dy_dz(), dtype=self.dtype))
[docs] @_setdoc_pro(reset_flow_control_initial_results.__doc__)
def set_theano_shared_structure(self, reset_ctrl=False):
"""
Set all theano shared variable dependent on :class:`Structure`.
Args:
reset_ctrl (bool): If true, [s0]
Returns:
True
"""
self.set_theano_shared_loop()
self.set_theano_shared_relations()
self.set_theano_shared_structure_surfaces()
# universal grades
# self.theano_graph.n_universal_eq_T.set_value(
# list(self.additional_data.kriging_data.df.loc['values', 'drift equations'].astype('int32')))
if reset_ctrl is True:
self.reset_flow_control_initial_results()
return True
def remove_series_without_data(self):
len_series_i = self.additional_data.structure_data.df.loc[
'values', 'len series surface_points'] - \
self.additional_data.structure_data.df.loc[
'values', 'number surfaces per series']
len_series_o = self.additional_data.structure_data.df.loc[
'values', 'len series orientations'].astype(
'int32')
# Remove series without data
non_zero_i = len_series_i.nonzero()[0]
non_zero_o = len_series_o.nonzero()[0]
non_zero = np.intersect1d(non_zero_i, non_zero_o)
self.non_zero = non_zero
return self.non_zero
def _compute_len_series(self):
self.len_series_i = self.additional_data.structure_data.df.loc[
'values', 'len series surface_points'] - \
self.additional_data.structure_data.df.loc[
'values', 'number surfaces per series']
self.len_series_o = self.additional_data.structure_data.df.loc[
'values', 'len series orientations'].astype(
'int32')
# Remove series without data
non_zero_i = self.len_series_i.nonzero()[0]
non_zero_o = self.len_series_o.nonzero()[0]
non_zero = np.intersect1d(non_zero_i, non_zero_o)
self.non_zero = non_zero
self.len_series_u = self.additional_data.kriging_data.df.loc[
'values', 'drift equations'].astype('int32')
try:
len_series_f_ = self.faults.faults_relations_df.values[non_zero][:, non_zero].sum(
axis=0)
except np.AxisError:
print('np.axis error')
len_series_f_ = self.faults.faults_relations_df.values.sum(axis=0)
self.len_series_f = np.atleast_1d(len_series_f_.astype(
'int32')) # [:self.additional_data.get_additional_data()['values']['Structure', 'number series']]
self._old_len_series = self.len_series_i
self.len_series_i = self.len_series_i[non_zero]
self.len_series_o = self.len_series_o[non_zero]
# self.len_series_f = self.len_series_f[non_zero]
self.len_series_u = self.len_series_u[non_zero]
if self.len_series_i.shape[0] == 0:
self.len_series_i = np.zeros(1, dtype=int)
self._old_len_series = self.len_series_i
if self.len_series_o.shape[0] == 0:
self.len_series_o = np.zeros(1, dtype=int)
if self.len_series_u.shape[0] == 0:
self.len_series_u = np.zeros(1, dtype=int)
if self.len_series_f.shape[0] == 0:
self.len_series_f = np.zeros(1, dtype=int)
self.len_series_w = self.len_series_i + self.len_series_o * 3 + self.len_series_u + self.len_series_f
[docs] def set_theano_shared_loop(self):
"""Set the theano shared variables that are looped for each series."""
self._compute_len_series()
self.theano_graph.len_series_i.set_value(
np.insert(self.len_series_i.cumsum(), 0, 0).astype('int32'))
self.theano_graph.len_series_o.set_value(
np.insert(self.len_series_o.cumsum(), 0, 0).astype('int32'))
self.theano_graph.len_series_w.set_value(
np.insert(self.len_series_w.cumsum(), 0, 0).astype('int32'))
# Number of surfaces per series. The function is not pretty but the result is quite clear
n_surfaces_per_serie = np.insert(
self.additional_data.structure_data.df.loc['values', 'number surfaces per series'][
self.non_zero].cumsum(), 0, 0). \
astype('int32')
self.theano_graph.n_surfaces_per_series.set_value(n_surfaces_per_serie)
self.theano_graph.n_universal_eq_T.set_value(
list(self.additional_data.kriging_data.df.loc['values', 'drift equations'].astype(
'int32')[self.non_zero]))
[docs] @_setdoc_pro(set_theano_shared_loop.__doc__)
def set_theano_shared_weights(self):
"""Set the theano shared weights and [s0]"""
self.set_theano_shared_loop()
self.theano_graph.weights_vector.set_value(
np.zeros((self.len_series_w.sum()), dtype=self.dtype))
def set_theano_shared_fault_relation(self):
self.remove_series_without_data()
"""Set the theano shared variable with the fault relation"""
self.theano_graph.fault_relation.set_value(
self.faults.faults_relations_df.values[self.non_zero][:, self.non_zero])
[docs] def set_theano_shared_is_fault(self):
"""Set theano shared variable which controls if a series is fault or not"""
is_fault_ = self.faults.df['isFault'].values[self.non_zero]
self.theano_graph.is_fault.set_value(is_fault_)
[docs] def set_theano_shared_is_finite(self):
"""Set theano shared variable which controls if a fault is finite or not"""
self.theano_graph.is_finite_ctrl.set_value(self.faults.df['isFinite'].values.astype(bool))
[docs] def set_theano_shared_onlap_erode(self):
"""Set the theano variables which control the masking patterns according to the uncomformity relation"""
self.remove_series_without_data()
is_erosion = self.series.df['BottomRelation'].values[self.non_zero] == 'Erosion'
is_onlap = np.roll(self.series.df['BottomRelation'].values[self.non_zero] == 'Onlap', 1)
if len(is_erosion) != 0:
is_erosion[-1] = False
# this comes from the series df
self.theano_graph.is_erosion.set_value(is_erosion)
self.theano_graph.is_onlap.set_value(is_onlap)
[docs] def set_theano_shared_faults(self):
"""Set all theano shared variables wich controls the faults behaviour"""
self.set_theano_shared_fault_relation()
# This comes from the faults df
self.set_theano_shared_is_fault()
self.set_theano_shared_is_finite()
[docs] def set_theano_shared_relations(self):
"""Set all theano shared variables that control all the series interactions with each other"""
self.set_theano_shared_fault_relation()
# This comes from the faults df
self.set_theano_shared_is_fault()
self.set_theano_shared_is_finite()
self.set_theano_shared_onlap_erode()
[docs] def set_initial_results(self):
"""
Initialize all the theano shared variables where we store the final results of the interpolation.
This function must be called always after set_theano_shared_loop
Returns:
True
"""
self._compute_len_series()
x_to_interp_shape = self.grid.values_c.shape[0] + 2 * self.len_series_i.sum()
n_series = self.len_series_i.shape[
0] # self.additional_data.structure_data.df.loc['values', 'number series']
self.theano_graph.weights_vector.set_value(
np.zeros((self.len_series_w.sum()), dtype=self.dtype))
self.theano_graph.scalar_fields_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype=self.dtype))
self.theano_graph.mask_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype='bool'))
self.theano_graph.block_matrix.set_value(
np.zeros(
(n_series, self.surfaces.df.iloc[:, self.surfaces._n_properties:].values.shape[1],
x_to_interp_shape), dtype=self.dtype))
return True
[docs] def set_initial_results_matrices(self):
"""
Initialize all the theano shared variables where we store the final results of the interpolation except the
kriging weights vector.
Returns:
True
"""
self._compute_len_series()
x_to_interp_shape = self.grid.values_c.shape[0] + 2 * self.len_series_i.sum()
n_series = self.len_series_i.shape[
0] # self.additional_data.structure_data.df.loc['values', 'number series']
self.theano_graph.scalar_fields_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype=self.dtype))
self.theano_graph.mask_matrix.set_value(
np.zeros((n_series, x_to_interp_shape), dtype='bool'))
self.theano_graph.block_matrix.set_value(
np.zeros(
(n_series, self.surfaces.df.iloc[:, self.surfaces._n_properties:].values.shape[1],
x_to_interp_shape), dtype=self.dtype))
def set_theano_shared_grid(self, grid=None):
if grid == 'shared':
grid_sh = self.grid.values_c
self.theano_graph.grid_val_T = theano.shared(grid_sh.astype(self.dtype),
'Constant values to interpolate.')
elif grid is not None:
self.theano_graph.grid_val_T = theano.shared(grid.astype(self.dtype),
'Constant values to interpolate.')
[docs] def modify_results_matrices_pro(self):
"""
Modify all theano shared matrices to the right size according to the structure data. This method allows
to change the size of the results without having the recompute all series"""
old_len_i = self._old_len_series
new_len_i = self.additional_data.structure_data.df.loc[
'values', 'len series surface_points'] - \
self.additional_data.structure_data.df.loc[
'values', 'number surfaces per series']
if new_len_i.shape[0] < old_len_i.shape[0]:
self.set_initial_results()
old_len_i = old_len_i[old_len_i != 0]
elif new_len_i.shape[0] > old_len_i.shape[0]:
self.set_initial_results()
new_len_i = new_len_i[new_len_i != 0]
else:
scalar_fields_matrix = self.theano_graph.scalar_fields_matrix.get_value()
mask_matrix = self.theano_graph.mask_matrix.get_value()
block_matrix = self.theano_graph.block_matrix.get_value()
len_i_diff = new_len_i - old_len_i
for e, i in enumerate(len_i_diff):
loc = self.grid.values_c.shape[0] + old_len_i[e]
i *= 2
if i == 0:
pass
elif i > 0:
self.theano_graph.scalar_fields_matrix.set_value(
np.insert(scalar_fields_matrix, [loc], np.zeros(i), axis=1))
self.theano_graph.mask_matrix.set_value(np.insert(
mask_matrix, [loc], np.zeros(i, dtype=self.dtype), axis=1))
self.theano_graph.block_matrix.set_value(np.insert(
block_matrix, [loc], np.zeros(i, dtype=self.dtype), axis=2))
else:
self.theano_graph.scalar_fields_matrix.set_value(
np.delete(scalar_fields_matrix, np.arange(loc, loc + i, -1) - 1, axis=1))
self.theano_graph.mask_matrix.set_value(
np.delete(mask_matrix, np.arange(loc, loc + i, -1) - 1, axis=1))
self.theano_graph.block_matrix.set_value(
np.delete(block_matrix, np.arange(loc, loc + i, -1) - 1, axis=2))
self.modify_results_weights()
[docs] def modify_results_weights(self):
"""Modify the theano shared weights vector according to the structure.
"""
old_len_w = self.len_series_w
self._compute_len_series()
new_len_w = self.len_series_w
if new_len_w.shape[0] != old_len_w[0]:
self.set_initial_results()
else:
weights = self.theano_graph.weights_vector.get_value()
len_w_diff = new_len_w - old_len_w
for e, i in enumerate(len_w_diff):
# print(len_w_diff, weights)
if i == 0:
pass
elif i > 0:
self.theano_graph.weights_vector.set_value(
np.insert(weights, old_len_w[e], np.zeros(i)))
else:
# print(np.delete(weights, np.arange(old_len_w[e], old_len_w[e] + i, -1)-1))
self.theano_graph.weights_vector.set_value(
np.delete(weights, np.arange(old_len_w[e], old_len_w[e] + i, -1) - 1))
[docs] def print_theano_shared(self):
"""Print many of the theano shared variables"""
print('len sereies i', self.theano_graph.len_series_i.get_value())
print('len sereies o', self.theano_graph.len_series_o.get_value())
print('len sereies w', self.theano_graph.len_series_w.get_value())
print('n surfaces per series', self.theano_graph.n_surfaces_per_series.get_value())
print('n universal eq', self.theano_graph.n_universal_eq_T.get_value())
print('is finite', self.theano_graph.is_finite_ctrl.get_value())
print('is erosion', self.theano_graph.is_erosion.get_value())
print('is onlap', self.theano_graph.is_onlap.get_value())
[docs] def compile_th_fn_geo(self, inplace=False, debug=True, grid: Union[str, np.ndarray] = None):
"""
Compile and create the theano function which can be evaluated to compute the geological models
Args:
inplace (bool): If true add the attribute theano.function to the object inplace
debug (bool): If true print some of the theano flags
grid: If None, grid will be passed as variable. If shared or np.ndarray the grid will be treated as
constant (if shared the grid will be taken of grid)
Returns:
theano.function: function that computes the whole interpolation
"""
self.set_all_shared_parameters(reset_ctrl=False)
# This are the shared parameters and the compilation of the function. This will be hidden as well at some point
input_data_T = self.theano_graph.input_parameters_loop
print('Compiling theano function...')
if grid == 'shared' or grid is not None:
self.set_theano_shared_grid(grid)
th_fn = theano.function(input_data_T,
self.theano_graph.theano_output(),
updates=[
(self.theano_graph.block_matrix, self.theano_graph.new_block),
(self.theano_graph.weights_vector,
self.theano_graph.new_weights),
(self.theano_graph.scalar_fields_matrix,
self.theano_graph.new_scalar),
(self.theano_graph.mask_matrix, self.theano_graph.new_mask)
],
on_unused_input='ignore',
allow_input_downcast=False,
profile=False)
if inplace is True:
self.theano_function = th_fn
if debug is True:
print('Level of Optimization: ', theano.config.optimizer)
print('Device: ', theano.config.device)
print('Precision: ', theano.config.floatX)
print('Number of faults: ',
self.additional_data.structure_data.df.loc['values', 'number faults'])
print('Compilation Done!')
return th_fn
