Source code for openquake.commonlib.writers

# -*- coding: utf-8 -*-
# vim: tabstop=4 shiftwidth=4 softtabstop=4
#
# Copyright (C) 2010-2017 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
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# 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 <http://www.gnu.org/licenses/>.

import re
import logging
import numpy  # this is needed by the doctests, don't remove it
from openquake.hazardlib import InvalidFile
from openquake.baselib.node import scientificformat

FIVEDIGITS = '%.5E'


[docs]class HeaderTranslator(object): r""" An utility to convert the headers in CSV files. When reading, the column names are converted into column descriptions with the method .read, when writing column descriptions are converted into column names with the method .write. The usage is >>> htranslator = HeaderTranslator( ... '(asset_ref):\|S100', ... '(eid):uint32', ... '(taxonomy):object') >>> htranslator.write('asset_ref:|S100 value:5'.split()) ['asset_ref', 'value:5'] >>> htranslator.read('asset_ref value:5'.split()) ['asset_ref:|S100', 'value:5'] """ def __init__(self, *regexps): self.suffix = [] short_regexps = [] for regex in regexps: prefix, suffix = regex.split(')') short_regexps.append(prefix + ')$') self.suffix.append(suffix) self.short_regex = '|'.join(short_regexps) self.long_regex = '|'.join(regexps)
[docs] def read(self, names): """ Convert names into descriptions """ descrs = [] for name in names: mo = re.match(self.short_regex, name) if mo: idx = mo.lastindex # matching group index, starting from 1 suffix = self.suffix[idx - 1].replace(r':\|', ':|') descrs.append(mo.group(mo.lastindex) + suffix + name[mo.end():]) else: descrs.append(name) return descrs
[docs] def write(self, descrs): """ Convert descriptions into names """ # example: '(poe-[\d\.]+):float32' -> 'poe-[\d\.]+' names = [] for descr in descrs: mo = re.match(self.long_regex, descr) if mo: names.append(mo.group(mo.lastindex) + descr[mo.end():]) else: names.append(descr) return names
htranslator = HeaderTranslator( '(rlzi):uint16', '(sid):uint32', '(eid):uint64', '(imti):uint8', '(gmv):float32', '(aid):uint32', '(boundary):object', '(tectonic_region_type):object', '(asset_ref):\|S100', '(rup_id):uint32', '(event_id):uint64', '(event_set):uint32', '(eid):uint32', '(year):uint32', '(taxonomy):\|S100', '(multiplicity):uint16', '(magnitude):float32', '(centroid_lon):float32', '(centroid_lat):float32', '(centroid_depth):float32', '(numsites):uint32', '(losses):float32', '(poes):float32', '(avg):float32', '(poe-[\d\.]+):float32', '(lon):float32', '(lat):float32', '(depth):float32', '(structural.*):float32', '(nonstructural.*):float32', '(business_interruption.*):float32', '(contents.*):float32', '(occupants):float32', '(occupants~.+):float32', '(occupants_ins):float32', '(no_damage):float32', '(slight):float32', '(moderate):float32', '(extensive):float32', '(complete):float32', '(\d+):float32', # realization column, used in the GMF scenario exporter ) # recursive function used internally by build_header def _build_header(dtype, root): header = [] if dtype.fields is None: if not root: return [] return [root + (str(dtype), dtype.shape)] for field in dtype.names: dt = dtype.fields[field][0] if dt.subdtype is None: # nested header.extend(_build_header(dt, root + (field,))) else: numpytype = str(dt.subdtype[0]) header.append(root + (field, numpytype, dt.shape)) return header # NB: builds an header that can be read by parse_header
[docs]def build_header(dtype): """ Convert a numpy nested dtype into a list of strings suitable as header of csv file. >>> imt_dt = numpy.dtype([('PGA', float, 3), ('PGV', float, 4)]) >>> build_header(imt_dt) ['PGA:3', 'PGV:4'] >>> gmf_dt = numpy.dtype([('A', imt_dt), ('B', imt_dt), ... ('idx', numpy.uint32)]) >>> build_header(gmf_dt) ['A~PGA:3', 'A~PGV:4', 'B~PGA:3', 'B~PGV:4', 'idx:uint32'] """ header = _build_header(dtype, ()) h = [] for col in header: name = '~'.join(col[:-2]) numpytype = col[-2] shape = col[-1] coldescr = name if numpytype != 'float64': coldescr += ':' + numpytype if shape: coldescr += ':' + ':'.join(map(str, shape)) h.append(coldescr) return h
[docs]def extract_from(data, fields): """ Extract data from numpy arrays with nested records. >>> imt_dt = numpy.dtype([('PGA', float, 3), ('PGV', float, 4)]) >>> a = numpy.array([([1, 2, 3], [4, 5, 6, 7])], imt_dt) >>> extract_from(a, ['PGA']) array([[ 1., 2., 3.]]) >>> gmf_dt = numpy.dtype([('A', imt_dt), ('B', imt_dt), ... ('idx', numpy.uint32)]) >>> b = numpy.array([(([1, 2, 3], [4, 5, 6, 7]), ... ([1, 2, 4], [3, 5, 6, 7]), 8)], gmf_dt) >>> extract_from(b, ['idx']) array([8], dtype=uint32) >>> extract_from(b, ['B', 'PGV']) array([[ 3., 5., 6., 7.]]) """ for f in fields: data = data[f] return data
[docs]def write_csv(dest, data, sep=',', fmt='%.6E', header=None, comment=None): """ :param dest: file, filename or io.StringIO instance :param data: array to save :param sep: separator to use (default comma) :param fmt: formatting string (default '%12.8E') :param header: optional list with the names of the columns to display :param comment: optional first line starting with a # character """ close = True if len(data) == 0: logging.warn('%s is empty', dest) if hasattr(dest, 'write'): # file-like object in append mode # it must be closed by client code close = False elif not hasattr(dest, 'getvalue'): # not a StringIO, assume dest is a filename dest = open(dest, 'w') try: # see if data is a composite numpy array data.dtype.fields except AttributeError: # not a composite array autoheader = [] else: autoheader = build_header(data.dtype) if comment: dest.write('# %s\n' % comment) someheader = header or autoheader if header != 'no-header' and someheader: dest.write(sep.join(htranslator.write(someheader)) + u'\n') if autoheader: all_fields = [col.split(':', 1)[0].split('~') for col in autoheader] for record in data: row = [] for fields in all_fields: val = extract_from(record, fields) if fields[0] in ('lon', 'lat', 'depth'): row.append('%.5f' % val) else: row.append(scientificformat(val, fmt)) dest.write(sep.join(row) + u'\n') else: for row in data: dest.write(sep.join(scientificformat(col, fmt) for col in row) + u'\n') if hasattr(dest, 'getvalue'): return dest.getvalue()[:-1] # a newline is strangely added elif close: dest.close() return dest.name
[docs]class CsvWriter(object): """ Class used in the exporters to save a bunch of CSV files """ def __init__(self, sep=',', fmt='%12.8E'): self.sep = sep self.fmt = fmt self.fnames = set()
[docs] def save(self, data, fname, header=None): """ Save data on fname. :param data: numpy array or list of lists :param fname: path name :param header: header to use """ write_csv(fname, data, self.sep, self.fmt, header) self.fnames.add(getattr(fname, 'name', fname))
[docs] def getsaved(self): """ Returns the list of files saved by this CsvWriter """ return sorted(self.fnames)
[docs]def castable_to_int(s): """ Return True if the string `s` can be interpreted as an integer """ try: int(s) except ValueError: return False else: return True
[docs]def parse_header(header): """ Convert a list of the form `['fieldname:fieldtype:fieldsize',...]` into a numpy composite dtype. The parser understands headers generated by :func:`openquake.commonlib.writers.build_header`. Here is an example: >>> parse_header(['PGA:float32', 'PGV', 'avg:float32:2']) (['PGA', 'PGV', 'avg'], dtype([('PGA', '<f4'), ('PGV', '<f8'), ('avg', '<f4', (2,))])) :params header: a list of type descriptions :returns: column names and the corresponding composite dtype """ triples = [] fields = [] for col_str in header: col = col_str.strip().split(':') n = len(col) if n == 1: # default dtype and no shape col = [col[0], 'float64', ''] elif n == 2: if castable_to_int(col[1]): # default dtype and shape col = [col[0], 'float64', col[1]] else: # dtype and no shape col = [col[0], col[1], ''] elif n > 3: raise ValueError('Invalid column description: %s' % col_str) field = col[0] numpytype = col[1] shape = () if not col[2].strip() else (int(col[2]),) triples.append((field, numpytype, shape)) fields.append(field) return fields, numpy.dtype(triples)
def _cast(col, ntype, shape, lineno, fname): # convert strings into tuples or numbers, used inside read_composite_array if shape: return tuple(map(ntype, col.split())) else: return ntype(col) # NB: this only works with flat composite arrays
[docs]def read_composite_array(fname, sep=','): r""" Convert a CSV file with header into a numpy array of records. >>> from openquake.baselib.general import writetmp >>> fname = writetmp('PGA:3,PGV:2,avg:1\n' ... '.1 .2 .3,.4 .5,.6\n') >>> print(read_composite_array(fname)) # array of shape (1,) [([0.1, 0.2, 0.3], [0.4, 0.5], [0.6])] """ with open(fname) as f: header = next(f) transheader = htranslator.read(header.split(sep)) fields, dtype = parse_header(transheader) ts_pairs = [] # [(type, shape), ...] for name in fields: dt = dtype.fields[name][0] ts_pairs.append((dt.subdtype[0].type if dt.subdtype else dt.type, dt.shape)) col_ids = list(range(1, len(ts_pairs) + 1)) num_columns = len(col_ids) records = [] col, col_id = '', 0 for i, line in enumerate(f, 2): row = line.split(sep) if len(row) != num_columns: raise InvalidFile( 'expected %d columns, found %d in file %s, line %d' % (num_columns, len(row), fname, i)) try: record = [] for (ntype, shape), col, col_id in zip(ts_pairs, row, col_ids): record.append(_cast(col, ntype, shape, i, fname)) records.append(tuple(record)) except Exception as e: raise InvalidFile( 'Could not cast %r in file %s, line %d, column %d ' 'using %s: %s' % (col, fname, i, col_id, (ntype.__name__,) + shape, e)) return numpy.array(records, dtype)
# this is simple and without error checking for the moment
[docs]def read_array(fname, sep=','): r""" Convert a CSV file without header into a numpy array of floats. >>> from openquake.baselib.general import writetmp >>> print(read_array(writetmp('.1 .2, .3 .4, .5 .6\n'))) [[[ 0.1 0.2] [ 0.3 0.4] [ 0.5 0.6]]] """ with open(fname) as f: records = [] for line in f: row = line.split(sep) record = [list(map(float, col.split())) for col in row] records.append(record) return numpy.array(records)
if __name__ == '__main__': # pretty print of NRML files import sys import shutil from openquake.hazardlib import nrml nrmlfiles = sys.argv[1:] for fname in nrmlfiles: node = nrml.read(fname) shutil.copy(fname, fname + '.bak') with open(fname, 'w') as out: nrml.write(list(node), out)