Source code for openquake.commonlib.logictree

# -*- coding: utf-8 -*-
# vim: tabstop=4 shiftwidth=4 softtabstop=4
# Copyright (C) 2010-2020 GEM Foundation
# OpenQuake is free software: you can redistribute it and/or modify it
# under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# OpenQuake is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# GNU Affero General Public License for more details.
# You should have received a copy of the GNU Affero General Public License
# along with OpenQuake. If not, see <>.

Logic tree parser, verifier and processor. See specs at

A logic tree object must be iterable and yielding realizations, i.e. objects
with attributes `value`, `weight`, `lt_path` and `ordinal`.

import io
import os
import re
import time
import logging
import functools
import itertools
import collections
import operator
from collections import namedtuple
import toml
import numpy
from openquake.baselib import hdf5
from openquake.baselib.python3compat import decode
from openquake.baselib.node import node_from_elem, Node as N, context
from openquake.baselib.general import (groupby, group_array, duplicated,
                                       add_defaults, AccumDict)
from openquake.hazardlib.gsim.mgmpe.avg_gmpe import AvgGMPE
from openquake.hazardlib.gsim.base import CoeffsTable
from openquake.hazardlib.imt import from_string
from openquake.hazardlib import valid, nrml, InvalidFile, pmf
from openquake.hazardlib.sourceconverter import SourceGroup
from import (
    Branch, BranchSet, LogicTreeError, parse_uncertainty, sample)

TRT_REGEX = re.compile(r'tectonicRegion="([^"]+?)"')
ID_REGEX = re.compile(r'id="([^"]+?)"')
SOURCE_TYPE_REGEX = re.compile(r'<(\w+Source)\b')

U16 = numpy.uint16
U32 = numpy.uint32
I32 = numpy.int32
F32 = numpy.float32

rlz_dt = numpy.dtype([
    ('ordinal', U32),
    ('branch_path', hdf5.vstr),
    ('weight', F32)

source_model_dt = numpy.dtype([
    ('name', hdf5.vstr),
    ('weight', F32),
    ('path', hdf5.vstr),
    ('samples', U32),
    ('offset', U32),

src_group_dt = numpy.dtype(
    [('grp_id', U32),
     ('name', hdf5.vstr),
     ('trti', U16),
     ('effrup', I32),
     ('totrup', I32),
     ('sm_id', U32)])

branch_dt = [('branchset', hdf5.vstr), ('branch', hdf5.vstr),
             ('utype', hdf5.vstr), ('uvalue', hdf5.vstr), ('weight', float)]

[docs]def unique(objects, key=None): """ Raise a ValueError if there is a duplicated object, otherwise returns the objects as they are. """ dupl = [] for obj, group in itertools.groupby(sorted(objects), key): if sum(1 for _ in group) > 1: dupl.append(obj) if dupl: raise ValueError('Found duplicates %s' % dupl) return objects
[docs]class Realization(object): """ Generic Realization object with attributes value, weight, ordinal, lt_path, samples and optionally offset. """ def __init__(self, value, weight, ordinal, lt_path, samples, offset=0): self.value = value self.weight = weight self.ordinal = ordinal self.lt_path = lt_path self.samples = samples self.offset = offset @property def pid(self): return '_'.join(self.lt_path) # path ID @property def name(self): """ Compact representation for the names """ names = self.value.split() if len(names) == 1: return names[0] elif len(names) == 2: return ' '.join(names) else: return ' '.join([names[0], '...', names[-1]]) def __repr__(self): samples = ', samples=%d' % self.samples if self.samples > 1 else '' return '<%s #%d %s, path=%s, weight=%s%s>' % ( self.__class__.__name__, self.ordinal, self.value, '_'.join(self.lt_path), self.weight, samples)
[docs]@functools.lru_cache() def get_effective_rlzs(rlzs): """ Group together realizations with the same path and yield the first representative of each group. """ effective = [] ordinal = 0 for group in groupby(rlzs, operator.attrgetter('pid')).values(): rlz = group[0] if all(path == '@' for path in rlz.lt_path): # empty realization continue effective.append( Realization(rlz.value, sum(r.weight for r in group), ordinal, rlz.lt_path, len(group))) ordinal += 1 return effective
# manage the legacy logicTreeBranchingLevel nodes def _bsnodes(fname, branchinglevel): if branchinglevel.tag.endswith('logicTreeBranchingLevel'): if len(branchinglevel) > 1: raise InvalidLogicTree( '%s: Branching level %s has multiple branchsets' % (fname, branchinglevel['branchingLevelID'])) return branchinglevel.nodes elif branchinglevel.tag.endswith('logicTreeBranchSet'): return [branchinglevel] else: raise ValueError('Expected BranchingLevel/BranchSet, got %s' % branchinglevel) Info = collections.namedtuple('Info', 'smpaths, applytosources')
[docs]def collect_info(smlt): """ Given a path to a source model logic tree, collect all of the path names to the source models it contains and build 1. a dictionary source model branch ID -> paths 2. a dictionary source model branch ID -> source IDs in applyToSources :param smlt: source model logic tree file :returns: an Info namedtupled containing the two dictionaries """ n = try: blevels = n.logicTree except Exception: raise InvalidFile('%s is not a valid source_model_logic_tree_file' % smlt) paths = set() applytosources = collections.defaultdict(list) # branchID -> source IDs for blevel in blevels: for bset in _bsnodes(smlt, blevel): if 'applyToSources' in bset.attrib: applytosources[bset.get('applyToBranches')].extend( bset['applyToSources'].split()) if bset['uncertaintyType'] in 'sourceModel extendModel': for br in bset: with context(smlt, br): fnames = unique(br.uncertaintyModel.text.split()) paths.update(_abs_paths(smlt, fnames)) return Info(sorted(paths), applytosources)
def _abs_paths(smlt, fnames): # relative -> absolute paths base_path = os.path.dirname(smlt) paths = [] for fname in fnames: if os.path.isabs(fname): raise InvalidFile('%s: %s must be a relative path' % (smlt, fname)) fname = os.path.abspath(os.path.join(base_path, fname)) if os.path.exists(fname): # consider only real paths paths.append(fname) return paths
[docs]def read_source_groups(fname): """ :param fname: a path to a source model XML file :return: a list of SourceGroup objects containing source nodes """ smodel = src_groups = [] if smodel[0].tag.endswith('sourceGroup'): # NRML 0.5 format for sg_node in smodel: sg = SourceGroup(sg_node['tectonicRegion']) sg.sources = sg_node.nodes src_groups.append(sg) else: # NRML 0.4 format: smodel is a list of source nodes src_groups.extend(SourceGroup.collect(smodel)) return src_groups
[docs]class SourceModelLogicTree(object): """ Source model logic tree parser. :param filename: Full pathname of logic tree file :raises LogicTreeError: If logic tree file has a logic error, which can not be prevented by xml schema rules (like referencing sources with missing id). """ _xmlschema = None FILTERS = ('applyToTectonicRegionType', 'applyToSources', 'applyToSourceType')
[docs] @classmethod def fake(cls): """ :returns: a fake SourceModelLogicTree with a single branch """ self = object.__new__(cls) arr = numpy.array([('bs0', 'b0', 'sourceModel', 'fake.xml', 1)], branch_dt) dic = dict(filename='fake.xml', seed=0, num_samples=0) self.__fromh5__(arr, dic) return self
def __init__(self, filename, seed=0, num_samples=0): self.filename = filename self.basepath = os.path.dirname(filename) self.seed = seed self.num_samples = num_samples self.branches = {} # branch_id -> branch self.bsetdict = {} self.previous_branches = [] self.tectonic_region_types = set() self.source_types = set() self.root_branchset = None root = try: tree = root.logicTree except AttributeError: raise LogicTreeError( root, self.filename, "missing logicTree node") self.parse_tree(tree) @property def on_each_source(self): """ True if there is an applyToSources for each source. """ return ( and == self.source_ids)
[docs] def parse_tree(self, tree_node): """ Parse the whole tree and point ``root_branchset`` attribute to the tree's root. """ = collect_info(self.filename) self.source_ids = collections.defaultdict(list) t0 = time.time() for depth, blnode in enumerate(tree_node.nodes): [bsnode] = _bsnodes(self.filename, blnode) self.parse_branchset(bsnode, depth) dt = time.time() - t0 bname = os.path.basename(self.filename)'Validated %s in %.2f seconds', bname, dt)
[docs] def parse_branchset(self, branchset_node, depth): """ :param branchset_ node: ``etree.Element`` object with tag "logicTreeBranchSet". :param depth: The sequential number of this branching level, based on 0. Enumerates children branchsets and call :meth:`parse_branchset`, :meth:`validate_branchset`, :meth:`parse_branches` and finally :meth:`apply_branchset` for each. Keeps track of "open ends" -- the set of branches that don't have any child branchset on this step of execution. After processing of every branchset only those branches that are listed in it can have child branchsets (if there is one on the next level). """ attrs = branchset_node.attrib.copy() self.bsetdict[attrs.pop('branchSetID')] = attrs uncertainty_type = branchset_node.attrib.get('uncertaintyType') filters = dict((filtername, branchset_node.attrib.get(filtername)) for filtername in self.FILTERS if filtername in branchset_node.attrib) self.validate_filters(branchset_node, uncertainty_type, filters) filters = self.parse_filters(branchset_node, uncertainty_type, filters) branchset = BranchSet(uncertainty_type, filters) self.validate_branchset(branchset_node, depth, branchset) self.parse_branches(branchset_node, branchset) if self.root_branchset is None: # not set yet self.num_paths = 1 self.root_branchset = branchset else: apply_to_branches = branchset_node.attrib.get('applyToBranches') if apply_to_branches: self.apply_branchset( apply_to_branches, branchset_node.lineno, branchset) else: for branch in self.previous_branches: branch.bset = branchset self.previous_branches = branchset.branches self.num_paths *= len(branchset.branches)
[docs] def parse_branches(self, branchset_node, branchset): """ Create and attach branches at ``branchset_node`` to ``branchset``. :param branchset_node: Same as for :meth:`parse_branchset`. :param branchset: An instance of :class:`BranchSet`. Checks that each branch has :meth:`valid <validate_uncertainty_value>` value, unique id and that all branches have total weight of 1.0. :return: ``None``, all branches are attached to provided branchset. """ bs_id = branchset_node['branchSetID'] weight_sum = 0 branches = branchset_node.nodes values = [] for branchnode in branches: weight = ~branchnode.uncertaintyWeight weight_sum += weight value_node = node_from_elem(branchnode.uncertaintyModel) if value_node.text is not None: values.append(value_node.text.strip()) if branchset.uncertainty_type in ('sourceModel', 'extendModel'): try: for fname in value_node.text.strip().split(): if fname.endswith(('.xml', '.nrml')): # except UCERF self.collect_source_model_data( branchnode['branchID'], fname) except Exception as exc: raise LogicTreeError( value_node, self.filename, str(exc)) from exc value = parse_uncertainty(branchset.uncertainty_type, value_node, self.filename) branch_id = branchnode.attrib.get('branchID') branch = Branch(bs_id, branch_id, weight, value) if branch_id in self.branches: raise LogicTreeError( branchnode, self.filename, "branchID '%s' is not unique" % branch_id) self.branches[branch_id] = branch branchset.branches.append(branch) if abs(weight_sum - 1.0) > pmf.PRECISION: raise LogicTreeError( branchset_node, self.filename, "branchset weights don't sum up to 1.0") if len(set(values)) < len(values): raise LogicTreeError( branchset_node, self.filename, "there are duplicate values in uncertaintyModel: " + ' '.join(values))
def __iter__(self): """ Yield Realization tuples. Notice that the weight is homogeneous when sampling is enabled, since it is accounted for in the sampling procedure. """ if self.num_samples: # random sampling of the logic tree weight = 1. / self.num_samples for i in range(self.num_samples): smlt_path = self.root_branchset.sample(self.seed + i) name = smlt_path[0].value smlt_path_ids = [branch.branch_id for branch in smlt_path] yield Realization(name, weight, None, tuple(smlt_path_ids), 1) else: # full enumeration ordinal = 0 for weight, smlt_path in self.root_branchset.enumerate_paths(): name = smlt_path[0].value smlt_branch_ids = [branch.branch_id for branch in smlt_path] yield Realization(name, weight, ordinal, tuple(smlt_branch_ids), 1) ordinal += 1
[docs] def parse_filters(self, branchset_node, uncertainty_type, filters): """ See superclass' method for description and signature specification. Converts "applyToSources" filter value by just splitting it to a list. """ if 'applyToSources' in filters: filters['applyToSources'] = filters['applyToSources'].split() return filters
[docs] def validate_filters(self, branchset_node, uncertainty_type, filters): """ See superclass' method for description and signature specification. Checks that the following conditions are met: * "sourceModel" uncertainties can not have filters. * Absolute uncertainties must have only one filter -- "applyToSources", with only one source id. * All other uncertainty types can have either no or one filter. * Filter "applyToSources" must mention only source ids that exist in source models. * Filter "applyToTectonicRegionType" must mention only tectonic region types that exist in source models. * Filter "applyToSourceType" must mention only source types that exist in source models. """ if uncertainty_type == 'sourceModel' and filters: raise LogicTreeError( branchset_node, self.filename, 'filters are not allowed on source model uncertainty') if len(filters) > 1: raise LogicTreeError( branchset_node, self.filename, "only one filter is allowed per branchset") if 'applyToTectonicRegionType' in filters: if not filters['applyToTectonicRegionType'] \ in self.tectonic_region_types: raise LogicTreeError( branchset_node, self.filename, "source models don't define sources of tectonic region " "type '%s'" % filters['applyToTectonicRegionType']) if uncertainty_type in ('abGRAbsolute', 'maxMagGRAbsolute', 'simpleFaultGeometryAbsolute', 'complexFaultGeometryAbsolute'): if not filters or not list(filters) == ['applyToSources'] \ or not len(filters['applyToSources'].split()) == 1: raise LogicTreeError( branchset_node, self.filename, "uncertainty of type '%s' must define 'applyToSources' " "with only one source id" % uncertainty_type) if uncertainty_type in ('simpleFaultDipRelative', 'simpleFaultDipAbsolute'): if not filters or (not ('applyToSources' in filters) and not ('applyToSourceType' in filters)): raise LogicTreeError( branchset_node, self.filename, "uncertainty of type '%s' must define either" "'applyToSources' or 'applyToSourceType'" % uncertainty_type) if 'applyToSourceType' in filters: if not filters['applyToSourceType'] in self.source_types: raise LogicTreeError( branchset_node, self.filename, "source models don't define sources of type '%s'" % filters['applyToSourceType']) if 'applyToSources' in filters: if (len(self.source_ids) > 1 and 'applyToBranches' not in branchset_node.attrib): raise LogicTreeError( branchset_node, self.filename, "applyToBranch must be " "specified together with applyToSources") for source_id in filters['applyToSources'].split(): cnt = sum(source_id in source_ids for source_ids in self.source_ids.values()) if cnt == 0: raise LogicTreeError( branchset_node, self.filename, "source with id '%s' is not defined in source " "models" % source_id)
[docs] def validate_branchset(self, branchset_node, depth, branchset): """ See superclass' method for description and signature specification. Checks that the following conditions are met: * First branching level must contain exactly one branchset, which must be of type "sourceModel". * All other branchsets must not be of type "sourceModel" or "gmpeModel". """ if depth == 0: if branchset.uncertainty_type != 'sourceModel': raise LogicTreeError( branchset_node, self.filename, 'first branchset must define an uncertainty ' 'of type "sourceModel"') else: if branchset.uncertainty_type == 'sourceModel': raise LogicTreeError( branchset_node, self.filename, 'uncertainty of type "sourceModel" can be defined ' 'on first branchset only') elif branchset.uncertainty_type == 'gmpeModel': raise LogicTreeError( branchset_node, self.filename, 'uncertainty of type "gmpeModel" is not allowed ' 'in source model logic tree')
[docs] def apply_branchset(self, apply_to_branches, lineno, branchset): """ See superclass' method for description and signature specification. Parses branchset node's attribute ``@applyToBranches`` to apply following branchests to preceding branches selectively. Branching level can have more than one branchset exactly for this: different branchsets can apply to different open ends. Checks that branchset tries to be applied only to branches on previous branching level which do not have a child branchset yet. """ for branch_id in apply_to_branches.split(): if branch_id not in self.branches: raise LogicTreeError( lineno, self.filename, "branch '%s' is not yet defined" % branch_id) branch = self.branches[branch_id] if branch.bset is not None: raise LogicTreeError( lineno, self.filename, "branch '%s' already has child branchset" % branch_id) branch.bset = branchset
def _get_source_model(self, source_model_file): # NB: do not remove this, it is meant to be overridden in the tests return open(os.path.join(self.basepath, source_model_file), encoding='utf-8')
[docs] def collect_source_model_data(self, branch_id, source_model): """ Parse source model file and collect information about source ids, source types and tectonic region types available in it. That information is used then for :meth:`validate_filters` and :meth:`validate_uncertainty_value`. """ # using regular expressions is a lot faster than using the with self._get_source_model(source_model) as sm: xml = self.tectonic_region_types.update(TRT_REGEX.findall(xml)) self.source_ids[branch_id].extend(ID_REGEX.findall(xml)) self.source_types.update(SOURCE_TYPE_REGEX.findall(xml))
[docs] def collapse(self, branchset_ids): """ Set the attribute .collapsed on the given branchsets """ for bsid, bset in self.bsetdict.items(): if bsid in branchset_ids: bset.collapsed = True
[docs] def bset_values(self, sm_rlz): """ :param sm_rlz: an effective realization :returns: a list of B - 1 pairs (branchset, value) """ return self.root_branchset.get_bset_values(sm_rlz.lt_path)[1:]
def _tomldict(self): out = {} for key, dic in self.bsetdict.items(): out[key] = toml.dumps({k: v.strip() for k, v in dic.items() if k != 'uncertaintyType'}).strip() return out def __toh5__(self): tbl = [] for brid, br in self.branches.items(): dic = self.bsetdict[br.bs_id].copy() utype = dic.pop('uncertaintyType') tbl.append((br.bs_id, brid, utype, br.value, br.weight)) attrs = self._tomldict() attrs['seed'] = self.seed attrs['num_samples'] = self.num_samples attrs['filename'] = self.filename return numpy.array(tbl, branch_dt), attrs def __fromh5__(self, array, attrs): bsets = [] self.branches = {} self.bsetdict = {} for bsid, rows in group_array(array, 'branchset').items(): utype = rows[0]['utype'] bset = BranchSet(utype, []) # TODO: filters = bsid for row in rows: br = Branch(bsid, row['branch'], row['weight'], row['uvalue']) self.branches[br.branch_id] = br bset.branches.append(br) bsets.append(bset) self.bsetdict[bsid] = {'uncertaintyType': utype} self.root_branchset = bsets[0] for i, childset in enumerate(bsets[1:]): dic = toml.loads(attrs[]) atb = dic.get('applyToBranches') for branch in bsets[i].branches: # parent branches if not atb or branch.branch_id in atb: branch.bset = bset self.seed = attrs['seed'] self.num_samples = attrs['num_samples'] self.filename = attrs['filename'] def __str__(self): return '<%s%s>' % (self.__class__.__name__, repr(self.root_branchset))
# used in GsimLogicTree BranchTuple = namedtuple('BranchTuple', 'trt id gsim weight effective')
[docs]class InvalidLogicTree(Exception): pass
[docs]class ImtWeight(object): """ A composite weight by IMTs extracted from the gsim_logic_tree_file """ def __init__(self, branch, fname): with context(fname, branch.uncertaintyWeight): nodes = list(branch.getnodes('uncertaintyWeight')) if 'imt' in nodes[0].attrib: raise InvalidLogicTree('The first uncertaintyWeight has an imt' ' attribute') self.dic = {'weight': float(nodes[0].text)} imts = [] for n in nodes[1:]: self.dic[n['imt']] = float(n.text) imts.append(n['imt']) if len(set(imts)) < len(imts): raise InvalidLogicTree( 'There are duplicated IMTs in the weights') def __mul__(self, other): new = object.__new__(self.__class__) if isinstance(other, self.__class__): keys = set(self.dic) | set(other.dic) new.dic = {k: self[k] * other[k] for k in keys} else: # assume a float new.dic = {k: self.dic[k] * other for k in self.dic} return new __rmul__ = __mul__ def __add__(self, other): new = object.__new__(self.__class__) if isinstance(other, self.__class__): new.dic = {k: self.dic[k] + other[k] for k in self.dic} else: # assume a float new.dic = {k: self.dic[k] + other for k in self.dic} return new __radd__ = __add__ def __truediv__(self, other): new = object.__new__(self.__class__) if isinstance(other, self.__class__): new.dic = {k: self.dic[k] / other[k] for k in self.dic} else: # assume a float new.dic = {k: self.dic[k] / other for k in self.dic} return new
[docs] def is_one(self): """ Check that all the inner weights are 1 up to the precision """ return all(abs(v - 1.) < pmf.PRECISION for v in self.dic.values() if v)
def __getitem__(self, imt): try: return self.dic[imt] except KeyError: return self.dic['weight'] def __repr__(self): return '<%s %s>' % (self.__class__.__name__, self.dic)
[docs]class GsimLogicTree(object): """ A GsimLogicTree instance is an iterable yielding `Realization` tuples with attributes `value`, `weight` and `lt_path`, where `value` is a dictionary {trt: gsim}, `weight` is a number in the interval 0..1 and `lt_path` is a tuple with the branch ids of the given realization. :param str fname: full path of the gsim_logic_tree file :param tectonic_region_types: a sequence of distinct tectonic region types :param ltnode: usually None, but it can also be a :class:`openquake.hazardlib.nrml.Node` object describing the GSIM logic tree XML file, to avoid reparsing it """
[docs] @classmethod def from_(cls, gsim): """ Generate a trivial GsimLogicTree from a single GSIM instance. """ ltbranch = N('logicTreeBranch', {'branchID': 'b1'}, nodes=[N('uncertaintyModel', text=str(gsim)), N('uncertaintyWeight', text='1.0')]) lt = N('logicTree', {'logicTreeID': 'lt1'}, nodes=[N('logicTreeBranchingLevel', {'branchingLevelID': 'bl1'}, nodes=[N('logicTreeBranchSet', {'applyToTectonicRegionType': '*', 'branchSetID': 'bs1', 'uncertaintyType': 'gmpeModel'}, nodes=[ltbranch])])]) return cls('fake/' + gsim.__class__.__name__, ['*'], ltnode=lt)
def __init__(self, fname, tectonic_region_types=['*'], ltnode=None): # tectonic_region_types usually comes from the source models self.filename = fname trts = sorted(tectonic_region_types) if len(trts) > len(set(trts)): raise ValueError( 'The given tectonic region types are not distinct: %s' % ','.join(trts)) self.values = collections.defaultdict(list) # {trt: gsims} self._ltnode = ltnode or self.bs_id_by_trt = {} self.branches = self._build_trts_branches(trts) # sorted by trt if trts != ['*']: # reduce self.values to the listed TRTs values = {} for trt in trts: values[trt] = self.values[trt] if not values[trt]: raise InvalidLogicTree('%s is missing the TRT %r' % (fname, trt)) self.values = values if trts and not self.branches: raise InvalidLogicTree( '%s is missing in %s' % (set(tectonic_region_types), fname)) @property def req_site_params(self): site_params = set() for trt in self.values: for gsim in self.values[trt]: site_params.update(gsim.REQUIRES_SITES_PARAMETERS) return site_params
[docs] def check_imts(self, imts): """ Make sure the IMTs are recognized by all GSIMs in the logic tree """ for trt in self.values: for gsim in self.values[trt]: for attr in dir(gsim): coeffs = getattr(gsim, attr) if not isinstance(coeffs, CoeffsTable): continue for imt in imts: if imt.startswith('SA'): try: coeffs[from_string(imt)] except KeyError: raise ValueError( '%s is out of the period range defined ' 'for %s' % (imt, gsim))
def __toh5__(self): weights = set() for branch in self.branches: weights.update(branch.weight.dic) dt = [('trt', hdf5.vstr), ('branch', hdf5.vstr), ('uncertainty', hdf5.vstr)] + [ (weight, float) for weight in sorted(weights)] branches = [(b.trt,, b.gsim) + tuple(b.weight[weight] for weight in sorted(weights)) for b in self.branches if b.effective] dic = {} if hasattr(self, 'filename'): # missing in EventBasedRiskTestCase case_1f dirname = os.path.dirname(self.filename) for gsims in self.values.values(): for gsim in gsims: for k, v in gsim.kwargs.items(): if k.endswith(('_file', '_table')): fname = os.path.join(dirname, v) with open(fname, 'rb') as f: dic[os.path.basename(v)] = return numpy.array(branches, dt), dic def __fromh5__(self, array, dic): self.branches = [] self.values = collections.defaultdict(list) for branch in array: br_id = branch['branch'] gsim = valid.gsim(branch['uncertainty']) for k, v in gsim.kwargs.items(): if k.endswith(('_file', '_table')): arr = numpy.asarray(dic[os.path.basename(v)][()]) gsim.kwargs[k] = io.BytesIO(bytes(arr)) gsim.__init__(**gsim.kwargs) self.values[branch['trt']].append(gsim) weight = object.__new__(ImtWeight) # branch has dtype ('trt', 'branch', 'uncertainty', 'weight', ...) weight.dic = {w: branch[w] for w in branch.dtype.names[3:]} if len(weight.dic) > 1: gsim.weight = weight bt = BranchTuple(branch['trt'], br_id, gsim, weight, True) self.branches.append(bt)
[docs] def reduce(self, trts): """ Reduce the GsimLogicTree. :param trts: a subset of tectonic region types :returns: a reduced GsimLogicTree instance """ new = object.__new__(self.__class__) vars(new).update(vars(self)) if trts != {'*'}: new.branches = [] for br in self.branches: branch = BranchTuple(br.trt,, br.gsim, br.weight, br.trt in trts) new.branches.append(branch) return new
[docs] def collapse(self, branchset_ids): """ Collapse the GsimLogicTree by using AgvGMPE instances if needed :param branchset_ids: branchset ids to collapse :returns: a collapse GsimLogicTree instance """ new = object.__new__(self.__class__) vars(new).update(vars(self)) new.branches = [] for trt, grp in itertools.groupby(self.branches, lambda b: b.trt): bs_id = self.bs_id_by_trt[trt] brs = [] gsims = [] weights = [] for br in grp: brs.append( gsims.append(br.gsim) weights.append(br.weight) if len(gsims) > 1 and bs_id in branchset_ids: kwargs = {} for brid, gsim, weight in zip(brs, gsims, weights): kw = gsim.kwargs.copy() kw['weight'] = weight.dic['weight'] kwargs[brid] = {gsim.__class__.__name__: kw} _toml = toml.dumps({'AvgGMPE': kwargs}) gsim = AvgGMPE(**kwargs) gsim._toml = _toml branch = BranchTuple(trt, bs_id, gsim, sum(weights), True) new.branches.append(branch) else: new.branches.append(br) return new
[docs] def get_num_branches(self): """ Return the number of effective branches for tectonic region type, as a dictionary. """ num = {} for trt, branches in itertools.groupby( self.branches, operator.attrgetter('trt')): num[trt] = sum(1 for br in branches if br.effective) return num
[docs] def get_num_paths(self): """ Return the effective number of paths in the tree. """ # NB: the algorithm assume a symmetric logic tree for the GSIMs; # in the future we may relax such assumption num_branches = self.get_num_branches() if not sum(num_branches.values()): return 0 num = 1 for val in num_branches.values(): if val: # the branch is effective num *= val return num
def _build_trts_branches(self, tectonic_region_types): # do the parsing, called at instantiation time to populate .values trts = [] branches = [] branchsetids = set() basedir = os.path.dirname(self.filename) for blnode in self._ltnode: [branchset] = _bsnodes(self.filename, blnode) if branchset['uncertaintyType'] != 'gmpeModel': raise InvalidLogicTree( '%s: only uncertainties of type "gmpeModel" ' 'are allowed in gmpe logic tree' % self.filename) bsid = branchset['branchSetID'] if bsid in branchsetids: raise InvalidLogicTree( '%s: Duplicated branchSetID %s' % (self.filename, bsid)) else: branchsetids.add(bsid) trt = branchset.get('applyToTectonicRegionType') if trt: # missing in self.bs_id_by_trt[trt] = bsid trts.append(trt) self.bs_id_by_trt[trt] = bsid # NB: '*' is used in scenario calculations to disable filtering effective = (tectonic_region_types == ['*'] or trt in tectonic_region_types) weights = [] branch_ids = [] for branch in branchset: weight = ImtWeight(branch, self.filename) weights.append(weight) branch_id = branch['branchID'] branch_ids.append(branch_id) try: gsim = valid.gsim(branch.uncertaintyModel, basedir) except Exception as exc: raise ValueError( "%s in file %s" % (exc, self.filename)) from exc if gsim in self.values[trt]: raise InvalidLogicTree('%s: duplicated gsim %s' % (self.filename, gsim)) if len(weight.dic) > 1: gsim.weight = weight self.values[trt].append(gsim) bt = BranchTuple( branchset['applyToTectonicRegionType'], branch_id, gsim, weight, effective) if effective: branches.append(bt) tot = sum(weights) assert tot.is_one(), '%s in branch %s' % (tot, branch_id) if duplicated(branch_ids): raise InvalidLogicTree( 'There where duplicated branchIDs in %s' % self.filename) if len(trts) > len(set(trts)): raise InvalidLogicTree( '%s: Found duplicated applyToTectonicRegionType=%s' % (self.filename, trts)) branches.sort(key=lambda b: (b.trt, # TODO: add an .idx to each GSIM ? return branches
[docs] def get_gsims(self, trt): """ :param trt: tectonic region type :returns: sorted list of available GSIMs for that trt """ if trt == '*' or trt == b'*': # fake logictree [trt] = self.values return sorted(self.values[trt])
[docs] def sample(self, n, seed): """ :param n: number of samples :param seed: random seed :returns: n Realization objects """ brlists = [sample([b for b in self.branches if b.trt == trt], n, seed + i) for i, trt in enumerate(self.values)] rlzs = [] for i in range(n): weight = 1 lt_path = [] lt_uid = [] value = [] for brlist in brlists: # there is branch list for each TRT branch = brlist[i] lt_path.append( lt_uid.append( if branch.effective else '@') weight *= branch.weight value.append(branch.gsim) rlz = Realization(tuple(value), weight, i, tuple(lt_uid), 1) rlzs.append(rlz) return rlzs
def __iter__(self): """ Yield :class:`openquake.commonlib.logictree.Realization` instances """ groups = [] # NB: branches are already sorted for trt in self.values: groups.append([b for b in self.branches if b.trt == trt]) # with T tectonic region types there are T groups and T branches for i, branches in enumerate(itertools.product(*groups)): weight = 1 lt_path = [] lt_uid = [] value = [] for trt, branch in zip(self.values, branches): lt_path.append( lt_uid.append( if branch.effective else '@') weight *= branch.weight value.append(branch.gsim) yield Realization(tuple(value), weight, i, tuple(lt_uid), 1) def __repr__(self): lines = ['%s,%s,%s,w=%s' % (b.trt,, b.gsim, b.weight['weight']) for b in self.branches if b.effective] return '<%s\n%s>' % (self.__class__.__name__, '\n'.join(lines))
[docs]def taxonomy_mapping(filename, taxonomies): """ :param filename: path to the CSV file containing the taxonomy associations :param taxonomies: an array taxonomy string -> taxonomy index :returns: (array, [[(taxonomy, weight), ...], ...]) """ if filename is None: # trivial mapping return (), [[(taxo, 1)] for taxo in taxonomies] dic = {} # taxonomy index -> risk taxonomy array = hdf5.read_csv(filename, {None: hdf5.vstr, 'weight': float}).array arr = add_defaults(array, weight=1.) assert arr.dtype.names == ('taxonomy', 'conversion', 'weight') dic = group_array(arr, 'taxonomy') taxonomies = taxonomies[1:] # strip '?' missing = set(taxonomies) - set(dic) if missing: raise InvalidFile('The taxonomies %s are in the exposure but not in %s' % (missing, filename)) lst = [[("?", 1)]] for idx, taxo in enumerate(taxonomies, 1): recs = dic[taxo] if abs(recs['weight'].sum() - 1.) > pmf.PRECISION: raise InvalidFile('%s: the weights do not sum up to 1 for %s' % (filename, taxo)) lst.append([(rec['conversion'], rec['weight']) for rec in recs]) return arr, lst
[docs]def capitalize(words): """ Capitalize words separated by spaces. """ return ' '.join(w.capitalize() for w in decode(words).split(' '))
[docs]def get_field(data, field, default): """ :param data: a record with a field `field`, possibily missing """ try: return data[field] except ValueError: # field missing in old engines return default
[docs]class LtRealization(object): """ Composite realization build on top of a source model realization and a GSIM realization. """ def __init__(self, ordinal, sm_lt_path, gsim_rlz, weight): self.ordinal = ordinal self.sm_lt_path = tuple(sm_lt_path) self.gsim_rlz = gsim_rlz self.weight = weight def __repr__(self): return '<%d,%s,w=%s>' % (self.ordinal,, self.weight) @property def gsim_lt_path(self): return self.gsim_rlz.lt_path @property def pid(self): """An unique identifier for effective realizations""" return '_'.join(self.sm_lt_path) + '~' + def __lt__(self, other): return self.ordinal < other.ordinal def __eq__(self, other): return repr(self) == repr(other) def __ne__(self, other): return repr(self) != repr(other) def __hash__(self): return hash(repr(self))
[docs]class FullLogicTree(object): """ The full logic tree as composition of :param source_model_lt: :class:`SourceModelLogicTree` object :param gsim_lt: :class:`GsimLogicTree` object """
[docs] @classmethod def fake(cls, gsimlt=None): """ :returns: a fake `FullLogicTree` instance with the given gsim logic tree object; if None, builds automatically a fake gsim logic tree """ gsim_lt = gsimlt or GsimLogicTree.from_('[FromFile]') fakeSM = Realization( 'scenario', weight=1, ordinal=0, lt_path='b1', samples=1) info = object.__new__(cls) info.source_model_lt = SourceModelLogicTree.fake() info.gsim_lt = gsim_lt info.sm_rlzs = [fakeSM] return info
def __init__(self, source_model_lt, gsim_lt): self.source_model_lt = source_model_lt self.gsim_lt = gsim_lt self.init() # set .sm_rlzs and .trt_by_grp
[docs] def init(self): # NB: the number of effective rlzs can be less than the number # of realizations in case of sampling sm_rlzs = get_effective_rlzs(self.source_model_lt) if not self.num_samples: num_gsim_rlzs = self.gsim_lt.get_num_paths() offset = 0 for sm_rlz in sm_rlzs: sm_rlz.offset = offset if self.num_samples: offset += sm_rlz.samples else: offset += num_gsim_rlzs self.sm_rlzs = sm_rlzs self.trti = {trt: i for i, trt in enumerate(self.gsim_lt.values)}
@property def trt_by_grp(self): """ :returns: a dictionary grp_id -> trt """ trt_by_grp = [] n = len(self.sm_rlzs) trts = list(self.gsim_lt.values) for smodel in self.sm_rlzs: for grp_id in self.grp_ids(smodel.ordinal): trt_by_grp.append((grp_id, trts[grp_id // n])) return dict(sorted(trt_by_grp)) @property def seed(self): """ :returns: the source_model_lt seed """ return self.source_model_lt.seed @property def num_samples(self): """ :returns: the source_model_lt ``num_samples`` parameter """ return self.source_model_lt.num_samples
[docs] def get_trti_eri(self, grp_id): """ :returns: (trti, eri) """ return divmod(grp_id, len(self.sm_rlzs))
[docs] def get_grp_id(self, trt, eri): """ :returns: grp_id """ return self.trti[trt] * len(self.sm_rlzs) + int(eri)
[docs] def grp_ids(self, eri): """ :param eri: effective realization index :returns: array of T group IDs, being T the number of TRTs """ nt = len(self.gsim_lt.values) ns = len(self.sm_rlzs) return eri + numpy.arange(nt) * ns
[docs] def get_samples_by_grp(self): """ :returns: a dictionary grp_id -> source_model.samples """ return {grp_id: sm.samples for sm in self.sm_rlzs for grp_id in self.grp_ids(sm.ordinal)}
[docs] def gsim_by_trt(self, rlz): """ :returns: a dictionary trt->gsim for the given realization """ return dict(zip(self.gsim_lt.values, rlz.gsim_rlz.value))
[docs] def get_rlzs(self, eri): """ :returns: a list of LtRealization objects """ rlzs = [] sm = self.sm_rlzs[eri] if self.num_samples: gsim_rlzs = self.gsim_lt.sample(sm.samples, self.seed + sm.ordinal) elif hasattr(self, 'gsim_rlzs'): # cache gsim_rlzs = self.gsim_rlzs else: self.gsim_rlzs = gsim_rlzs = get_effective_rlzs(self.gsim_lt) for i, gsim_rlz in enumerate(gsim_rlzs): weight = sm.weight * gsim_rlz.weight rlz = LtRealization(sm.offset + i, sm.lt_path, gsim_rlz, weight) rlzs.append(rlz) return rlzs
[docs] def get_realizations(self): """ :returns: the complete list of LtRealizations """ rlzs = sum((self.get_rlzs(sm.ordinal) for sm in self.sm_rlzs), []) assert rlzs, 'No realizations found??' if self.num_samples: assert len(rlzs) == self.num_samples, (len(rlzs), self.num_samples) for rlz in rlzs: for k in rlz.weight.dic: rlz.weight.dic[k] = 1. / self.num_samples else: tot_weight = sum(rlz.weight for rlz in rlzs) if not tot_weight.is_one(): # this may happen for rounding errors; we ensure the sum of # the weights is 1 for rlz in rlzs: rlz.weight = rlz.weight / tot_weight return rlzs
[docs] def get_rlzs_by_gsim(self, grp_id): """ :returns: a dictionary gsim -> rlzs """ trti, eri = divmod(grp_id, len(self.sm_rlzs)) rlzs_by_gsim = AccumDict(accum=[]) for rlz in self.get_rlzs(eri): rlzs_by_gsim[rlz.gsim_rlz.value[trti]].append(rlz.ordinal) return {gsim: U32(rlzs) for gsim, rlzs in sorted(rlzs_by_gsim.items())}
[docs] def get_rlzs_by_gsim_grp(self): """ :returns: a dictionary grp_id -> gsim -> rlzs """ dic = {} for sm in self.sm_rlzs: for grp_id in self.grp_ids(sm.ordinal): dic[grp_id] = self.get_rlzs_by_gsim(grp_id) return dic
[docs] def get_rlzs_by_grp(self): """ :returns: a dictionary grp_id -> [rlzis, ...] """ dic = {} for sm in self.sm_rlzs: for grp_id in self.grp_ids(sm.ordinal): grp = 'grp-%02d' % grp_id dic[grp] = list(self.get_rlzs_by_gsim(grp_id).values()) return dic # grp_id -> lists of rlzi
def __getnewargs__(self): # with this FullLogicTree instances will be unpickled correctly return self.seed, self.num_samples, self.sm_rlzs def __toh5__(self): # save full_lt/sm_data in the datastore sm_data = [] for sm in self.sm_rlzs: sm_data.append((sm.value, sm.weight, '_'.join(sm.lt_path), sm.samples, sm.offset)) return (dict( source_model_lt=self.source_model_lt, gsim_lt=self.gsim_lt, sm_data=numpy.array(sm_data, source_model_dt)), dict(seed=self.seed, num_samples=self.num_samples, trts=hdf5.array_of_vstr(self.gsim_lt.values))) def __fromh5__(self, dic, attrs): # TODO: this is called more times than needed, maybe we should cache it sm_data = dic['sm_data'] vars(self).update(attrs) self.source_model_lt = dic['source_model_lt'] self.gsim_lt = dic['gsim_lt'] self.sm_rlzs = [] for sm_id, rec in enumerate(sm_data): path = tuple(str(decode(rec['path'])).split('_')) sm = Realization( rec['name'], rec['weight'], sm_id, path, rec['samples'], rec['offset']) self.sm_rlzs.append(sm)
[docs] def get_num_rlzs(self, sm_rlz=None): """ :param sm_rlz: a Realization instance (or None) :returns: the number of realizations per source model (or all) """ if sm_rlz is None: return sum(self.get_num_rlzs(sm) for sm in self.sm_rlzs) if self.num_samples: return sm_rlz.samples return self.gsim_lt.get_num_paths()
@property def rlzs(self): """ :returns: an array of realizations """ tups = [(r.ordinal,, r.weight['weight']) for r in self.get_realizations()] return numpy.array(tups, rlz_dt)
[docs] def get_gsims_by_trt(self): """ :returns: a dictionary trt -> sorted gsims """ if self.num_samples: gsims_by_trt = AccumDict(accum=set()) for sm in self.sm_rlzs: rlzs = self.gsim_lt.sample(sm.samples, self.seed + sm.ordinal) for t, trt in enumerate(self.gsim_lt.values): gsims_by_trt[trt].update([rlz.value[t] for rlz in rlzs]) else: gsims_by_trt = self.gsim_lt.values return {trt: sorted(gsims) for trt, gsims in gsims_by_trt.items()}
[docs] def get_sm_by_grp(self): """ :returns: a dictionary grp_id -> sm_id """ return {grp_id: sm.ordinal for sm in self.sm_rlzs for grp_id in self.grp_ids(sm.ordinal)}
def __repr__(self): info_by_model = {} for sm in self.sm_rlzs: info_by_model[sm.lt_path] = ( '_'.join(map(decode, sm.lt_path)), decode(sm.value), sm.weight, self.get_num_rlzs(sm)) summary = ['%s, %s, weight=%s: %d realization(s)' % ibm for ibm in info_by_model.values()] return '<%s\n%s>' % (self.__class__.__name__, '\n'.join(summary))