|
import copy
import time
import logging
import numpy as num
from pyrocko.guts import Int, Float, Bool
from pyrocko import gf
from grond.meta import Forbidden, GrondError
from ..base import Analyser, AnalyserConfig, AnalyserResult
logger = logging.getLogger('grond.analysers.target_balancer')
guts_prefix = 'grond'
class TargetBalancingAnalyser(Analyser):
""" Estimating target weights that balance the signal amplitudes.
Signal amplitudes depend on the source-receiver distance, on the
phase type and the taper used. Large signals have in general
a higher contribution to the misfit than smaller signals,
without carrying more information. With this function, weights are
estimated that shall balance the phase contributions.
The weight estimation is based on synthetic waveforms that stem from
a given number of random forward models. The inverse of the mean
synthetic signal amplitudes gives the balancing weight. This is
described as adaptive station weighting in Heimann (2011).
"""
def __init__(self, niter, use_reference_magnitude, cutoff):
Analyser.__init__(self)
self.niter = niter
self.use_reference_magnitude = use_reference_magnitude
self.cutoff = cutoff
def log_progress(self, problem, iiter, niter):
t = time.time()
if self._tlog_last < t - 10. \
or iiter == 0 \
or iiter == niter - 1:
logger.info(
'Target balancing for "%s" at %i/%i.' % (
problem.name,
iiter, niter))
self._tlog_last = t
def analyse(self, problem, ds):
if self.niter == 0:
return
wtargets = []
if not problem.has_waveforms:
return
for target in problem.waveform_targets:
wtarget = copy.copy(target)
wtarget.flip_norm = True
wtarget.weight = 1.0
wtargets.append(wtarget)
wproblem = problem.copy()
wproblem.targets = wtargets
xbounds = wproblem.get_parameter_bounds()
misfits = num.zeros((self.niter, wproblem.ntargets, 2))
rstate = num.random.RandomState(123)
isbad_mask = None
self._tlog_last = 0
for iiter in range(self.niter):
self.log_progress(problem, iiter, self.niter)
while True:
if self.use_reference_magnitude:
try:
fixed_magnitude = wproblem.base_source.get_magnitude()
except gf.DerivedMagnitudeError:
raise GrondError(
'Cannot use use_reference_magnitude for this type '
'of source model.')
else:
fixed_magnitude = None
x = wproblem.random_uniform(
xbounds, rstate, fixed_magnitude=fixed_magnitude)
try:
x = wproblem.preconstrain(x)
break
except Forbidden:
pass
if isbad_mask is not None and num.any(isbad_mask):
isok_mask = num.logical_not(isbad_mask)
else:
isok_mask = None
misfits[iiter, :, :] = wproblem.misfits(x, mask=isok_mask)
isbad_mask = num.isnan(misfits[iiter, :, 1])
mean_ms = num.mean(misfits[:, :, 0], axis=0)
mean_ps = num.mean(misfits[:, :, 1], axis=0)
weights = 1. / mean_ps
families, nfamilies = wproblem.get_family_mask()
for ifamily in range(nfamilies):
weights[families == ifamily] /= (
num.nansum(weights[families == ifamily]) /
num.nansum(num.isfinite(weights[families == ifamily])))
if self.cutoff is not None:
weights[mean_ms / mean_ps > self.cutoff] = 0.0
for weight, target in zip(weights, problem.waveform_targets):
target.analyser_results['target_balancing'] = \
TargetBalancingAnalyserResult(weight=float(weight))
for itarget, target in enumerate(problem.waveform_targets):
logger.info((
'Balancing analysis for target "%s":\n'
' m/p: %g\n'
' weight: %g\n'
) % (
target.string_id(),
mean_ms[itarget] / mean_ps[itarget],
weights[itarget]))
class TargetBalancingAnalyserResult(AnalyserResult):
weight = Float.T()
class TargetBalancingAnalyserConfig(AnalyserConfig):
"""Configuration parameters of the target balancing."""
niterations = Int.T(
default=1000,
help='Number of random forward models for mean phase amplitude '
'estimation')
use_reference_magnitude = Bool.T(
default=False,
help='Fix magnitude of random sources to the magnitude of the '
'reference event.')
cutoff = Float.T(
optional=True,
help='Remove targets where ratio m/p > cutoff, where m is the misfit '
'between synthetics and observations and p is the misfit between '
'synthetics and zero-traces. Magnitude should be fixed to use '
'this.')
def get_analyser(self):
return TargetBalancingAnalyser(
niter=self.niterations,
use_reference_magnitude=self.use_reference_magnitude,
cutoff=self.cutoff)
__all__ = '''
TargetBalancingAnalyser
TargetBalancingAnalyserConfig
'''.split()
|