"""
This file is part of CLIMADA.
Copyright (C) 2017 ETH Zurich, CLIMADA contributors listed in AUTHORS.
CLIMADA is free software: you can redistribute it and/or modify it under the
terms of the GNU General Public License as published by the Free
Software Foundation, version 3.
CLIMADA is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with CLIMADA. If not, see <https://www.gnu.org/licenses/>.
---
Define ImpactCalc class.
"""
__all__ = ['ImpactCalc']
import logging
import numpy as np
from scipy import sparse
import geopandas as gpd
from climada import CONFIG
from climada.engine import Impact
LOGGER = logging.getLogger(__name__)
[docs]class ImpactCalc():
"""
Class to compute impacts from exposures, impact function set and hazard
"""
[docs] def __init__(self,
exposures,
impfset,
hazard):
"""
ImpactCalc constructor
The dimension of the imp_mat variable must be compatible with the
exposures and hazard objects.
Parameters
----------
exposures : climada.entity.Exposures
exposures used to compute impacts
impf_set: climada.entity.ImpactFuncSet
impact functions set used to compute impacts
hazard : climada.Hazard
hazard used to compute impacts
"""
self.exposures = exposures
self.impfset = impfset
self.hazard = hazard
# exposures index to use for matrix reconstruction
self._orig_exp_idx = np.arange(self.exposures.gdf.shape[0])
@property
def n_exp_pnt(self):
"""Number of exposure points (rows in gdf)"""
return self.exposures.gdf.shape[0]
@property
def n_events(self):
"""Number of hazard events (size of event_id array)"""
return self.hazard.size
[docs] def impact(self, save_mat=True, assign_centroids=True,
ignore_cover=False, ignore_deductible=False):
"""Compute the impact of a hazard on exposures.
Parameters
----------
save_mat : bool, optional
if true, save the total impact matrix (events x exposures)
Default: True
assign_centroids : bool, optional
indicates whether centroids are assigned to the self.exposures object.
Centroids assignment is an expensive operation; set this to ``False`` to save
computation time if the hazards' centroids are already assigned to the exposures
object.
Default: True
ignore_cover : bool, optional
if set to True, the column 'cover' of the exposures GeoDataFrame, if present, is
ignored and the impact it not capped by the values in this column.
Default: False
ignore_deductible : bool, opotional
if set to True, the column 'deductible' of the exposures GeoDataFrame, if present, is
ignored and the impact it not reduced through values in this column.
Default: False
Examples
--------
>>> haz = Hazard.from_mat(HAZ_DEMO_MAT) # Set hazard
>>> impfset = ImpactFuncSet.from_excel(ENT_TEMPLATE_XLS)
>>> exp = Exposures(pd.read_excel(ENT_TEMPLATE_XLS))
>>> impcalc = ImpactCal(exp, impfset, haz)
>>> imp = impcalc.impact(insured=True)
>>> imp.aai_agg
See also
--------
apply_deductible_to_mat : apply deductible to impact matrix
apply_cover_to_mat : apply cover to impact matrix
"""
# check for compability of exposures and hazard type
if all(name not in self.exposures.gdf.columns for
name in ['if_', f'if_{self.hazard.haz_type}',
'impf_', f'impf_{self.hazard.haz_type}']):
raise AttributeError(
"Impact calculation not possible. No impact functions found "
f"for hazard type {self.hazard.haz_type} in exposures."
)
# check for compability of impact function and hazard type
if not self.impfset.get_func(haz_type=self.hazard.haz_type):
raise AttributeError(
"Impact calculation not possible. No impact functions found "
f"for hazard type {self.hazard.haz_type} in impf_set."
)
impf_col = self.exposures.get_impf_column(self.hazard.haz_type)
exp_gdf = self.minimal_exp_gdf(impf_col, assign_centroids, ignore_cover, ignore_deductible)
if exp_gdf.size == 0:
return self._return_empty(save_mat)
LOGGER.info('Calculating impact for %s assets (>0) and %s events.',
exp_gdf.size, self.n_events)
imp_mat_gen = self.imp_mat_gen(exp_gdf, impf_col)
insured = ('cover' in exp_gdf and exp_gdf.cover.max() >= 0) \
or ('deductible' in exp_gdf and exp_gdf.deductible.max() > 0)
if insured:
LOGGER.info("cover and/or deductible columns detected,"
" going to calculate insured impact")
#TODO: make a better impact matrix generator for insured impacts when
# the impact matrix is already present
imp_mat_gen = self.insured_mat_gen(imp_mat_gen, exp_gdf, impf_col)
return self._return_impact(imp_mat_gen, save_mat)
def _return_impact(self, imp_mat_gen, save_mat):
"""Return an impact object from an impact matrix generator
Parameters
----------
imp_mat_gen : generator
Generator of impact matrix and corresponding exposures index
save_mat : boolean
if true, save the impact matrix
Returns
-------
Impact
Impact Object initialize from the impact matrix
See Also
--------
imp_mat_gen : impact matrix generator
insured_mat_gen: insured impact matrix generator
"""
if save_mat:
imp_mat = self.stitch_impact_matrix(imp_mat_gen)
at_event, eai_exp, aai_agg = \
self.risk_metrics(imp_mat, self.hazard.frequency)
else:
imp_mat = None
at_event, eai_exp, aai_agg = self.stitch_risk_metrics(imp_mat_gen)
return Impact.from_eih(
self.exposures, self.hazard, at_event, eai_exp, aai_agg, imp_mat
)
def _return_empty(self, save_mat):
"""
Return empty impact.
Parameters
----------
save_mat : bool
If true, save impact matrix
Returns
-------
Impact
Empty impact object with correct array sizes.
"""
at_event = np.zeros(self.n_events)
eai_exp = np.zeros(self.n_exp_pnt)
aai_agg = 0.0
if save_mat:
imp_mat = sparse.csr_matrix((
self.n_events, self.n_exp_pnt), dtype=np.float64
)
else:
imp_mat = None
return Impact.from_eih(
self.exposures, self.hazard, at_event, eai_exp, aai_agg, imp_mat
)
[docs] def minimal_exp_gdf(self, impf_col, assign_centroids, ignore_cover, ignore_deductible):
"""Get minimal exposures geodataframe for impact computation
Parameters
----------
exposures : climada.entity.Exposures
hazard : climada.Hazard
impf_col : str
Name of the impact function column in exposures.gdf
assign_centroids : bool
Indicates whether centroids are re-assigned to the self.exposures object
or kept from previous impact calculation with a hazard of the same hazard type.
Centroids assignment is an expensive operation; set this to ``False`` to save
computation time if the centroids have not changed since the last impact
calculation.
include_cover : bool
if set to True, the column 'cover' of the exposures GeoDataFrame is excluded from the
returned GeoDataFrame, otherwise it is included if present.
include_deductible : bool
if set to True, the column 'deductible' of the exposures GeoDataFrame is excluded from
the returned GeoDataFrame, otherwise it is included if present.
"""
if assign_centroids:
self.exposures.assign_centroids(self.hazard, overwrite=True)
elif self.hazard.centr_exp_col not in self.exposures.gdf.columns:
raise ValueError("'assign_centroids' is set to 'False' but no centroids are assigned"
f" for the given hazard type ({self.hazard.haz_type})."
" Run 'exposures.assign_centroids()' beforehand or set"
" 'assign_centroids' to 'True'")
mask = (
(self.exposures.gdf.value.values == self.exposures.gdf.value.values) # value != NaN
& (self.exposures.gdf.value.values != 0) # value != 0
& (self.exposures.gdf[self.hazard.centr_exp_col].values >= 0) # centroid assigned
)
columns = ['value', impf_col, self.hazard.centr_exp_col]
if not ignore_cover and 'cover' in self.exposures.gdf:
columns.append('cover')
if not ignore_deductible and 'deductible' in self.exposures.gdf:
columns.append('deductible')
exp_gdf = gpd.GeoDataFrame(
{col: self.exposures.gdf[col].values[mask]
for col in columns},
)
if exp_gdf.size == 0:
LOGGER.warning("No exposures with value >0 in the vicinity of the hazard.")
self._orig_exp_idx = mask.nonzero()[0] # update index of kept exposures points in exp_gdf
# within the full exposures
return exp_gdf
[docs] def imp_mat_gen(self, exp_gdf, impf_col):
"""
Generator of impact sub-matrices and correspoding exposures indices
The exposures gdf is decomposed into chunks that fit into the max
defined memory size. For each chunk, the impact matrix is computed
and returned, together with the corresponding exposures points index.
Parameters
----------
exp_gdf : GeoDataFrame
Geodataframe of the exposures with columns required for impact
computation.
impf_col : str
name of the desired impact column in the exposures.
Raises
------
ValueError
if the hazard is larger than the memory limit
Yields
------
scipy.sparse.crs_matrix, np.ndarray
impact matrix and corresponding exposures indices for each chunk.
"""
def _chunk_exp_idx(haz_size, idx_exp_impf):
'''
Chunk computations in sizes that roughly fit into memory
'''
max_size = CONFIG.max_matrix_size.int()
if haz_size > max_size:
raise ValueError(
f"Hazard size '{haz_size}' exceeds maximum matrix size '{max_size}'. "
"Increase max_matrix_size configuration parameter accordingly."
)
n_chunks = np.ceil(haz_size * len(idx_exp_impf) / max_size)
return np.array_split(idx_exp_impf, n_chunks)
for impf_id in exp_gdf[impf_col].dropna().unique():
impf = self.impfset.get_func(
haz_type=self.hazard.haz_type, fun_id=impf_id
)
idx_exp_impf = (exp_gdf[impf_col].values == impf_id).nonzero()[0]
for exp_idx in _chunk_exp_idx(self.hazard.size, idx_exp_impf):
exp_values = exp_gdf.value.values[exp_idx]
cent_idx = exp_gdf[self.hazard.centr_exp_col].values[exp_idx]
yield (
self.impact_matrix(exp_values, cent_idx, impf),
exp_idx
)
[docs] def insured_mat_gen(self, imp_mat_gen, exp_gdf, impf_col):
"""
Generator of insured impact sub-matrices (with applied cover and deductible)
and corresponding exposures indices
This generator takes a 'regular' impact matrix generator and applies cover and
deductible onto the impacts. It yields the same sub-matrices as the original
generator.
Deductible and cover are taken from the dataframe stored in `exposures.gdf`.
Parameters
----------
imp_mat_gen : generator of tuples (sparse.csr_matrix, np.array)
The generator for creating the impact matrix. It returns a part of the full
matrix and the associated exposure indices.
exp_gdf : GeoDataFrame
Geodataframe of the exposures with columns required for impact computation.
impf_col : str
Name of the column in 'exp_gdf' indicating the impact function (id)
Yields
------
mat : scipy.sparse.csr_matrix
Impact sub-matrix (with applied cover and deductible) with size
(n_events, len(exp_idx))
exp_idx : np.array
Exposure indices for impacts in mat
"""
for mat, exp_idx in imp_mat_gen:
impf_id = exp_gdf[impf_col][exp_idx[0]]
cent_idx = exp_gdf[self.hazard.centr_exp_col].values[exp_idx]
impf = self.impfset.get_func(
haz_type=self.hazard.haz_type,
fun_id=impf_id)
if 'deductible' in exp_gdf:
deductible = exp_gdf.deductible.values[exp_idx]
mat = self.apply_deductible_to_mat(mat, deductible, self.hazard, cent_idx, impf)
if 'cover' in exp_gdf:
cover = exp_gdf.cover.values[exp_idx]
mat = self.apply_cover_to_mat(mat, cover)
yield (mat, exp_idx)
[docs] def impact_matrix(self, exp_values, cent_idx, impf):
"""
Compute the impact matrix for given exposure values,
assigned centroids, a hazard, and one impact function.
Parameters
----------
exp_values : np.array
Exposure values
cent_idx : np.array
Hazard centroids assigned to each exposure location
hazard : climada.Hazard
Hazard object
impf : climada.entity.ImpactFunc
one impactfunction comon to all exposure elements in exp_gdf
Returns
-------
scipy.sparse.csr_matrix
Impact per event (rows) per exposure point (columns)
"""
n_exp_pnt = len(cent_idx) # implicitly checks in matrix assignement whether
# len(cent_idx) == len(exp_values)
mdr = self.hazard.get_mdr(cent_idx, impf)
exp_values_csr = sparse.csr_matrix( # vector 1 x exp_size
(exp_values, np.arange(n_exp_pnt), [0, n_exp_pnt]),
shape=(1, n_exp_pnt))
fract = self.hazard._get_fraction(cent_idx) # pylint: disable=protected-access
if fract is None:
return mdr.multiply(exp_values_csr)
return fract.multiply(mdr).multiply(exp_values_csr)
[docs] def stitch_impact_matrix(self, imp_mat_gen):
"""
Make an impact matrix from an impact sub-matrix generator
"""
# rows: events index
# cols: exposure point index within self.exposures
data, row, col = np.hstack([
(mat.data, mat.nonzero()[0], self._orig_exp_idx[idx][mat.nonzero()[1]])
for mat, idx in imp_mat_gen
])
return sparse.csr_matrix(
(data, (row, col)), shape=(self.n_events, self.n_exp_pnt)
)
[docs] def stitch_risk_metrics(self, imp_mat_gen):
"""Compute the impact metrics from an impact sub-matrix generator
This method is used to compute the risk metrics if the user decided not to store
the full impact matrix.
Parameters
----------
imp_mat_gen : generator of tuples (sparse.csr_matrix, np.array)
The generator for creating the impact matrix. It returns a part of the full
matrix and the associated exposure indices.
Returns
-------
at_event : np.array
Accumulated damage for each event
eai_exp : np.array
Expected impact within a period of 1/frequency_unit for each exposure point
aai_agg : float
Average impact within a period of 1/frequency_unit aggregated
"""
at_event = np.zeros(self.n_events)
eai_exp = np.zeros(self.n_exp_pnt)
for sub_imp_mat, idx in imp_mat_gen:
at_event += self.at_event_from_mat(sub_imp_mat)
eai_exp[self._orig_exp_idx[idx]] += \
self.eai_exp_from_mat(sub_imp_mat, self.hazard.frequency)
aai_agg = self.aai_agg_from_eai_exp(eai_exp)
return at_event, eai_exp, aai_agg
[docs] @staticmethod
def apply_deductible_to_mat(mat, deductible, hazard, cent_idx, impf):
"""
Apply a deductible per exposure point to an impact matrix at given
centroid points for given impact function.
All exposure points must have the same impact function. For different
impact functions apply use this method repeatedly on the same impact
matrix.
Parameters
----------
imp_mat : scipy.sparse.csr_matrix
impact matrix (events x exposure points)
deductible : np.array()
deductible for each exposure point
hazard : climada.Hazard
hazard used to compute the imp_mat
cent_idx : np.array()
index of centroids associated with each exposure point
impf : climada.entity.ImpactFunc
impact function associated with the exposure points
Returns
-------
imp_mat : scipy.sparse.csr_matrix
impact matrix with applied deductible
"""
paa = hazard.get_paa(cent_idx, impf)
mat -= paa.multiply(sparse.csr_matrix(deductible))
mat.eliminate_zeros()
return mat
[docs] @staticmethod
def apply_cover_to_mat(mat, cover):
"""
Apply cover to impact matrix.
The impact data is clipped to the range [0, cover]. The cover is defined
per exposure point.
Parameters
----------
imp_mat : scipy.sparse.csr_matrix
impact matrix
cover : np.array()
cover per exposures point (columns of imp_mat)
Returns
-------
imp_mat : scipy.sparse.csr_matrix
impact matrix with applied cover
"""
mat.data = np.clip(mat.data, 0, cover[mat.nonzero()[1]])
mat.eliminate_zeros()
return mat
[docs] @staticmethod
def eai_exp_from_mat(mat, freq):
"""
Compute impact for each exposures from the total impact matrix
Parameters
----------
imp_mat : sparse.csr_matrix
matrix num_events x num_exp with impacts.
frequency : np.array
frequency of events within a period of 1/frequency_unit
Returns
-------
eai_exp : np.array
expected impact within a period of 1/frequency_unit for each exposure
"""
n_events = freq.size
freq_csr = sparse.csr_matrix( #vector n_events x 1
(freq, np.zeros(n_events), np.arange(n_events + 1)),
shape=(n_events, 1))
return mat.multiply(freq_csr).sum(axis=0).A1
[docs] @staticmethod
def at_event_from_mat(mat):
"""
Compute impact for each hazard event from the total impact matrix
Parameters
----------
imp_mat : sparse.csr_matrix
matrix num_events x num_exp with impacts.
Returns
-------
at_event : np.array
impact for each hazard event
"""
return np.asarray(np.sum(mat, axis=1)).ravel()
[docs] @staticmethod
def aai_agg_from_eai_exp(eai_exp):
"""
Aggregate impact.eai_exp
Parameters
----------
eai_exp : np.array
expected impact within a period of 1/frequency_unit for each exposure point
Returns
-------
float
average aggregated impact within a period of 1/frequency_unit
"""
return np.sum(eai_exp)
[docs] @classmethod
def risk_metrics(cls, mat, freq):
"""
Compute risk metricss eai_exp, at_event, aai_agg
for an impact matrix and a frequency vector.
Parameters
----------
mat : sparse.csr_matrix
matrix num_events x num_exp with impacts.
freq : np.array
array with the frequency per event
Returns
-------
eai_exp: np.array
expected impact within a period of 1/frequency_unit at each exposure point
at_event: np.array()
total impact for each event
aai_agg : float
average impact within a period of 1/frequency_unit aggregated over all exposure points
"""
eai_exp = cls.eai_exp_from_mat(mat, freq)
at_event = cls.at_event_from_mat(mat)
aai_agg = cls.aai_agg_from_eai_exp(eai_exp)
return at_event, eai_exp, aai_agg
