def work(self, search=True, forward=True, run_id='current'):
self.pre_work(search or forward)
seis = self.seismosizer
conf = self.in_config.get_config()
mm_conf = self.in_config.get_config(keys=Step.outer_misfit_method_params)
sourcetype = self.sourcetype
base_source = source_model.Source( sourcetype )
for p in source_model.param_names(sourcetype):
if d2u(p) in conf:
base_source[p] = float(conf[d2u(p)])
if 'plane' in conf and conf['plane'] == 2:
strike, dip, slip_rake = float(conf['strike']), float(conf['dip']), float(conf['slip_rake'])
strike, dip, slip_rake = moment_tensor.other_plane( strike, dip, slip_rake )
base_source['strike'] = strike
base_source['dip'] = dip
base_source['slip-rake'] = slip_rake
if 'plane' in conf:
for param in 'strike', 'dip', 'slip-rake':
self.out_config.__dict__['active_'+d2u(param)] = base_source[param]
grid_def = []
for param in self.params:
oldval = base_source[u2d(param)]
descr = conf[param+'_range']
grid_def.append(grid_defi(u2d(param),oldval,descr))
if search or forward: self.setup_inner_misfit_method()
if search:
if self.ref_source_from is not None:
step, ident = self.ref_source_from
ref_source = step.get_snapshot_source(ident)
else:
ref_source = None
self.setup_inner_misfit_method()
finder = gridsearch.MisfitGrid( base_source, param_values=grid_def, ref_source=ref_source)
finder.compute(seis)
else:
finder = self.load(self.stepname, run_id=run_id)
finder.postprocess(**mm_conf)
self.dump(finder, self.stepname)
stats = finder.stats
for param in self.params:
str_result = stats[u2d(param)].str_best_and_confidence()
logging.info(str_result)
self.result(str_result, param )
base_source[u2d(param)] = stats[u2d(param)].best
self.out_config.__dict__[param] = stats[u2d(param)].best
self.out_config.__dict__[param+'_stats'] = stats[u2d(param)]
self.out_config.min_misfit = finder.get_best_misfit()
self.out_config.nstations_total = finder.nreceivers
self.out_config.nstations_used = finder.nreceivers_enabled
logging.info('Misfit = %f, Source = %s' % (finder.get_best_misfit(), str(base_source)))
mt = base_source.moment_tensor()
logging.info(str(mt))
self.result(str(mt), 'moment_tensor')
misfit_median = finder.get_median_of_misfits_by_r()
if self.xblacklist_level is not None:
ir = 0
if 'xblacklist' in conf:
xblacklist = set(conf['xblacklist'])
else:
xblacklist = set()
for r, mm in zip(seis.receivers, finder.misfits_by_r):
if mm/misfit_median > self.xblacklist_level:
xblacklist.add(ir)
logging.info('Blacklisting: %i, %s, %g' % (ir+1, r.name, mm/misfit_median))
ir += 1
self.out_config.xblacklist = sorted(list(xblacklist))
if forward:
self.snapshot( base_source, 'best', mm_conf )
self.post_work(search or forward)
def work(self, search=True, forward=True, run_id='current'):
self.pre_work(search or forward)
seis = self.seismosizer
conf = self.in_config.get_config()
mm_conf = self.in_config.get_config(keys=Step.outer_misfit_method_params)
sourcetype = self.sourcetype
base_source = source_model.Source( sourcetype )
for p in source_model.param_names(sourcetype):
if d2u(p) in conf:
base_source[p] = float(conf[d2u(p)])
if 'plane' in conf and conf['plane'] == 2:
strike, dip, slip_rake = float(conf['strike']), float(conf['dip']), float(conf['slip_rake'])
strike, dip, slip_rake = moment_tensor.other_plane( strike, dip, slip_rake )
base_source['strike'] = strike
base_source['dip'] = dip
base_source['slip-rake'] = slip_rake
if 'plane' in conf:
for param in 'strike', 'dip', 'slip-rake':
self.out_config.__dict__['active_'+d2u(param)] = base_source[param]
# setup grid of starting points for gradient se
starter_grid_def = []
for param in self.params:
if param+'_start_range' in conf:
oldval = base_source[u2d(param)]
descr = conf[param+'_start_range']
starter_grid_def.append( grid_defi(u2d(param),oldval,descr) )
if starter_grid_def:
starter_sources = base_source.grid( starter_grid_def )
else:
starter_sources = [ base_source ]
grid_def = []
for param in self.params:
oldval = base_source[u2d(param)]
descr = conf[param+'_range']
grid_def.append(grid_defi(u2d(param),oldval,descr))
dparams = [ u2d(param) for param in self.params ]
self.nminfunccalls = 0
norms = [ num.min(vals[1:]-vals[:-1]) for (param, vals) in grid_def ]
bounds = [ (num.min(vals)/norm,num.max(vals)/norm) for ((param, vals),norm) in zip(grid_def, norms) ]
if search or forward: self.setup_inner_misfit_method()
def x_to_source(x):
source = copy.deepcopy(base_source)
for i,param in enumerate(self.params):
source[u2d(param)] = x[i]*norms[i]
return source
def source_to_x(source):
x = []
for param, norm in zip(self.params, norms):
x.append(source[u2d(param)]/norm)
return x
def minfunc(x):
self.nminfunccalls += 1
source = x_to_source(x)
try:
logging.debug('Evaluating source: %s', source.pretty_str(dparams) )
best, misfit = seis.best_source([source], **mm_conf)
logging.debug('Misfit: %g' % misfit)
except NoValidSources:
logging.warn('Gradient search aborted at invalid source:' )
for str_param in current_base.pretty_str(dparams).splitlines():
logging.warn( str_param )
raise
return misfit
if search:
self.setup_inner_misfit_method()
# fix depth range by trying out different depths
for iparam, (param, vals) in enumerate(grid_def):
if param == 'depth':
depth_sources = base_source.grid( [ (param,vals) ] )
dummy_source, dummy_misfit, failings = seis.best_source(depth_sources, return_failings=True, **mm_conf)
ok = []
for isource, source in enumerate(depth_sources):
if isource not in failings:
ok.append(source['depth'])
miok = min(ok)
maok = max(ok)
bounds[iparam] = ((miok+gridsearch.step_at(ok,miok)*0.3)/norms[iparam],
(maok-gridsearch.step_at(ok,maok)*0.3)/norms[iparam])
# grid search over gradient searches
min_misfit = None
very_best_source = None
ngood = 0
ntotal = 0
for starter_source in starter_sources:
ntotal += 1
# look if starting source is valid
current_base = copy.deepcopy(starter_source)
try:
current_base, misfit = seis.best_source([current_base], **mm_conf)
except NoValidSources:
logging.warn('Skipping invalid starting source:')
for str_param in current_base.pretty_str(dparams).splitlines():
logging.warn( str_param )
continue
if min_misfit == None:
min_misfit = misfit
very_best_source = current_base
x0 = source_to_x(current_base)
try:
x, misfit, d = fmin_l_bfgs_b(minfunc, x0, approx_grad=True, bounds=bounds, epsilon=0.2, factr=1e10)
except NoValidSources:
continue
current = x_to_source(x)
if d['warnflag'] != 0: continue
# restart gradient search at current minimum
x0 = source_to_x(current)
try:
x, misfit, d = fmin_l_bfgs_b(minfunc, x0, approx_grad=True, bounds=bounds, epsilon=0.05, factr=1e7)
except NoValidSources:
continue
current = x_to_source(x)
if d['warnflag'] != 0: continue
minkind = ''
if misfit < min_misfit:
logging.info('Found possible minimum: Misfit = %f' % misfit)
for str_param in current.pretty_str(dparams).splitlines():
logging.info( str_param )
if 'strike' in self.params:
mt = moment_tensor.MomentTensor(strike=current['strike'], dip=current['dip'], rake=current['slip-rake'])
fps1, fps2 = mt.str_fault_planes().strip().splitlines()
logging.info(fps1)
logging.info(fps2)
min_misfit = misfit
very_best_source = current
ngood += 1
if min_misfit is None:
raise Exception('No valid starting points found')
else:
logging.error('Fixme: inversion.py')
very_best_source_normalform = copy.deepcopy(very_best_source)
if 'strike' in base_source.keys():
very_best_source_normalform.disambigue_sdr()
for param, coords in grid_def:
val = very_best_source[param]
remark = ''
if snap(val,coords)[0] in (0,len(coords)-1): remark = ' (?: at end of range)'
str_result = '%s = %g%s' % (param.title(), val, remark)
logging.info(str_result)
self.result(str_result, d2u(param))
base_source[param] = very_best_source_normalform[param]
self.out_config.__dict__[d2u(param)] = very_best_source_normalform[param]
str_result = 'Misfit = %g' % min_misfit
logging.info(str_result)
self.result(str_result, 'Misfit')
self.out_config.__dict__['min_misfit'] = min_misfit
logging.info('Total number iterations in gradient searches: %i' % self.nminfunccalls)
logging.info('Number of gradient searches tried: %i' % ntotal)
logging.info('Number of successful gradient searches: %i' % ngood)
self.out_config.__dict__['greeper_nminfunccalls'] = self.nminfunccalls
self.out_config.__dict__['greeper_ntotal'] = ntotal
self.out_config.__dict__['greeper_ngood'] = ngood
if forward:
self.snapshot( base_source, 'best', mm_conf )
self.post_work(search or forward)
