'''
Manages the file format and class for correlations.
:copyright:
The SeisMIC development team (makus@gfz-potsdam.de).
:license:
EUROPEAN UNION PUBLIC LICENCE v. 1.2
(https://joinup.ec.europa.eu/collection/eupl/eupl-text-eupl-12)
:author:
Peter Makus (makus@gfz-potsdam.de)
Created: Friday, 16th April 2021 03:21:30 pm
Last Modified: Wednesday, 2nd October 2024 11:28:48 am
'''
import ast
import fnmatch
import os
import re
from typing import List
import warnings
from copy import deepcopy
import numpy as np
# from numpy.core.fromnumeric import compress
from obspy.core.utcdatetime import UTCDateTime
from obspy.core import Stats
import h5py
from seismic.correlate.stream import CorrStream, CorrTrace
import seismic.utils.miic_utils as mu
hierarchy = '/{tag}/{network}/{station}/{location}/' \
+ '{channel}/{corr_st}/{corr_et}'
h5_FMTSTR = os.path.join(
"{dir}", "{network}.{station}.{location}.{channel}.h5")
[docs]class DBHandler(h5py.File):
"""
The actual file handler of the hdf5 correlation files.
.. warning::
**Should not be accessed directly. Access
:class:`~seismic.db.corr_hdf5.CorrelationDataBase` instead.**
Child object of :class:`h5py.File` and inherets all its attributes and
functions in addition to functions that are particularly useful for noise
correlations.
"""
def __init__(self, path, mode, compression, co, force):
super(DBHandler, self).__init__(path, mode=mode)
if isinstance(compression, str):
self.compression = re.findall(r'(\w+?)(\d+)', compression)[0][0]
if self.compression != 'gzip':
raise ValueError(
'Compression of type %s is not supported.'
% self.compression)
self.compression_opts = int(
re.findall(r'(\w+?)(\d+)', compression)[0][1])
if self.compression_opts not in np.arange(1, 10, 1, dtype=int):
ii = np.argmin(abs(
np.arange(1, 10, 1, dtype=int) - self.compression_opts))
self.compression_opts = np.arange(1, 10, 1, dtype=int)[ii]
warnings.warn(
'Chosen compression level is not available for %s. \
%s Has been chosen instead (closest)' % (
self.compression, str(self.compression_opts)))
else:
self.compression = None
self.compression_opts = None
# check out the processing that was done on the correlations. Different
# data will not be allowed
if co is not None:
try:
co_old = self.get_corr_options()
if co_old != co_to_hdf5(co) and not force:
try:
diff = {k: (v, co_old[k]) for k, v in co_to_hdf5(
co).items() if v != co_old[k]}
except KeyError as e:
raise PermissionError(
f'One option is not defined in new dict. {e}'
)
raise PermissionError(
f'The output file {path} already exists and contains'
' data with'
' different processing parameters. Differences are:'
'\nFirst: New parameters; Second: Old parameters'
f'\n{diff}'
)
except KeyError:
self.add_corr_options(co)
def _close(self):
self.close()
[docs] def add_corr_options(self, co: dict):
sco = str(co_to_hdf5(co))
ds = self.create_dataset('co', data=np.empty(1))
ds.attrs['co'] = sco
[docs] def add_correlation(
self, data: CorrTrace or CorrStream, tag='subdivision'):
"""
Add correlation data to the hdf5 file. Can be accessed using the
:func:`~seismic.db.corr_hdf5.DBHandler.get_data()` method.
:param data: Data to save. Either a
:class:`~seismic.correlate.correlate.CorrTrace` object or a
:class:`~seismic.correlate.correlate.CorrStream` holding one or
several traces.
:type data: CorrTrace or CorrStream
:param tag: The tag that the data should be saved under. By convention,
unstacked correlations are saved with the tag `'subdivision'`,
whereas stacks are saved with the tag `stack_$stacklen$`, where
$stacklen$ is to be replaced by the length of the stack in seconds.
:raises TypeError: for wrong data type.
"""
if not isinstance(data, CorrTrace) and\
not isinstance(data, CorrStream):
raise TypeError('Data has to be either a \
:class:`~seismic.correlate.correlate.CorrTrace` object or a \
:class:`~seismic.correlate.correlate.CorrStream` object')
if isinstance(data, CorrTrace):
data = [data]
for tr in data:
st = tr.stats
path = hierarchy.format(
tag=tag,
network=st.network, station=st.station, channel=st.channel,
location=st.location,
corr_st=st.corr_start.format_fissures(),
corr_et=st.corr_end.format_fissures())
try:
ds = self.create_dataset(
path, data=tr.data, compression=self.compression,
compression_opts=self.compression_opts)
convert_header_to_hdf5(ds, st)
except ValueError as e:
print(e)
warnings.warn("The dataset %s is already in file and will be \
omitted." % path, category=UserWarning)
[docs] def remove_data(
self, network: str, station: str, location: str, channel: str,
tag: str, corr_start: UTCDateTime | str):
"""
Deletes the correlation from the file.
:param network: network comb string
:type network: str
:param station: station comb string
:type station: str
:param location: location combination code
:type location: str
:param channel: channel comb string
:type channel: str
:param tag: tag
:type tag: str
:param corr_start: Correlation Start, may either be a UTCDateTime
object or a string. String is allowed to contain wildcards.
:type corr_start: UTCDateTime | str
:raises TypeError: if corr_start is not string or UTCDateTime
"""
if '*' in network+station+channel+location:
raise ValueError(
'Network, Station, location, and channel '
'code may not contain wildcards.'
)
# Make sure the request is structured correctly
network, station, location, channel \
= mu.sort_combinations_alphabetically(
network, station, location, channel)
if isinstance(corr_start, UTCDateTime):
corr_start = corr_start.format_fissures()
elif not isinstance(corr_start, str):
raise TypeError(
'corr_start has to be either a string, in which case it '
'has to be given in fissures format, or a UTCDateTime object.'
f' Is {type(corr_start)}.'
)
path = hierarchy.format(
tag=tag, network=network, station=station, channel=channel,
location=location, corr_st=corr_start, corr_et='*')
while path[-2:] == '/*':
path = path[:-2]
# Extremely ugly way of changing the path
if '*' not in path and '?' not in path:
try:
del self[path]
except KeyError:
warnings.warn(f'requested dataset {path} not found')
return
# Now, we need to differ between the fnmatch pattern and the actually
# accessed path
path = path.replace('?', '*')
pattern = path.replace('/*', '*')
for sst in self.get_available_starttimes(
network, station, tag, location, channel)[channel]:
subpath = hierarchy.format(
tag=tag, network=network, station=station, channel=channel,
location=location,
corr_st=sst, corr_et='')[:-1]
if fnmatch.fnmatch(subpath, pattern):
del self[subpath]
[docs] def get_corr_options(self) -> dict:
try:
sco = str(self['co'].attrs['co'])
# Run once more through co_to_hdf5 to account for
# correlations that have been computed with older versions
co = co_to_hdf5(ast.literal_eval(sco))
except KeyError:
raise KeyError('No correlation options in file')
return co
[docs] def get_data(
self, network: str, station: str, location: str, channel: str,
tag: str, corr_start: UTCDateTime = None,
corr_end: UTCDateTime = None) -> CorrStream:
"""
Returns a :class:`~seismic.correlate.correlate.CorrStream` holding
all the requested data.
.. note::
Wildcards are allowed for all parameters.
:param network: network (combination), e.g., IU-YP
:type network: str
:param station: station (combination), e.g., HRV-BRK
:type station: str
:param channel: channel (combination), e.g., BZ-BR
:type channel: str
:param corr_start: starttime of the time windows used to computed this
correlation, defaults to None
:type corr_start: UTCDateTime, optional
:param corr_end: endtime of the time windows used to computed this
correlation, defaults to None
:type corr_end: UTCDateTime, optional
:return: a :class:`~seismic.correlate.correlate.CorrStream` holding
all the requested data.
:rtype: CorrStream
"""
# Make sure the request is structured correctly
network, station, location, channel =\
mu.sort_combinations_alphabetically(
network, station, location, channel)
if isinstance(corr_start, UTCDateTime):
corr_start = corr_start.format_fissures()
else:
try:
corr_start = UTCDateTime(corr_start).format_fissures()
except (ValueError, TypeError):
warnings.warn(
'Correlation start is not a valid UTCDateTime object. '
'Set to *')
corr_start = '*'
if isinstance(corr_end, UTCDateTime):
corr_end = corr_end.format_fissures()
else:
try:
corr_end = UTCDateTime(corr_end).format_fissures()
except (ValueError, TypeError):
warnings.warn(
'Correlation end is not a valid UTCDateTime object. '
'Set to *')
corr_end = '*'
path = hierarchy.format(
tag=tag, network=network, station=station, channel=channel,
location=location,
corr_st=corr_start, corr_et=corr_end)
# Extremely ugly way of changing the path
if '*' not in path and '?' not in path:
data = np.array(self[path])
header = read_hdf5_header(self[path])
return CorrStream(CorrTrace(data, _header=header))
# Now, we need to differ between the fnmatch pattern and the actually
# accessed path
path = path.replace('?', '*')
pattern = path.replace('/*', '*')
path = path.split('*')[0]
return all_traces_recursive(self[path], CorrStream(), pattern)
[docs] def get_available_starttimes(
self, network: str, station: str, tag: str, location: str,
channel: str or list = '*') -> dict:
"""
Returns a dictionary with channel codes as keys and available
correlation starttimes as values.
..note::
Wildcards are only allowed for channel.
:param network: Network code (combined code, e.g., IU-YP)
:type network: str
:param station: Station code (combined)
:type station: str
:param tag: Tag
:type tag: str
:param location: Combined Location code.
:type location: str
:param channel: Channel code (combined), wildcards allowed,
defaults to '*'
:type channel: str, optional
:return: A dictionary holding the availabe starttimes for each channel
combination
:rtype: dict
"""
path = hierarchy.format(
tag=tag, network=network, station=station, channel=channel,
location=location,
corr_st='*', corr_et='*')
out = {}
if isinstance(channel, str):
if '*' not in channel:
path = '/'.join(path.split('/')[:-2])
try:
out[channel] = list(self[path].keys())
except KeyError:
pass
return out
channel = [channel]
path = '/'.join(path.split('/')[:-3])
for ch in channel:
for match in fnmatch.filter(self[path].keys(), ch):
out[match] = list(self['/'.join([path, match])].keys())
return out
[docs] def get_available_channels(
self, tag: str, network: str, station: str,
location: str) -> List[str]:
"""
Returns a list of all available channels (i.e., their combination
codes) in this file.
:param tag: The tag that this data was save as.
:type tag: str
:param network: Network combination code in the form `net0-net1`
:type network: str
:param station: Network combination code in the form `stat0-stat1`
:type station: str
:param location: Location combination code
:type location: str
:return: A list of all channel combinations.
:rtype: List[str]
"""
path = hierarchy.format(
tag=tag, network=network, station=station, location=location,
channel='*', corr_st='*', corr_et='*')
path = path.split('*')[0]
try:
return list(self[path].keys())
except KeyError:
# No channels available
return []
[docs]class CorrelationDataBase(object):
"""
Base class to handle the hdf5 files that contain noise correlations.
"""
def __init__(
self, path: str, corr_options: dict = None, mode: str = 'a',
compression: str = 'gzip3', _force: bool = False):
"""
Access an hdf5 file holding correlations. The resulting file can be
accessed using all functionalities of
`h5py <https://www.h5py.org/>`_ (for example as a dict).
:param path: Full path to the file
:type path: str
:param corr_options: The dictionary holding the parameters for the
correlation. If set to `None`. The mode will be set to read-only
`= 'r'`. Defaults to None.
:type corr_options: dict or None
:param mode: Mode to access the file. Options are: 'a' for all, 'w' for
write, 'r+' for writing in an already existing file, or 'r' for
read-only , defaults to 'a'.
:type mode: str, optional
:param compression: The compression algorithm and compression level
that the arrays should be saved with. 'gzip3' tends to perform
well, else you could choose 'gzipx' where x is a digit between
1 and 9 (i.e., 9 is the highest compression) or None for fastest
perfomance, defaults to 'gzip3'.
:type compression: str, optional
:param _force: allow differnt correlation options, defaults to False
:type _force: bool, optional
.. warning::
**Access only through a context manager (see below):**
>>> with CorrelationDataBase(myfile.h5) as cdb:
>>> type(cdb) # This is a DBHandler
<class 'seismic.db.corr_hdf5.DBHandler'>
Example::
>>> with CorrelationDataBase(
'/path/to/db/XN-XN.NEP06-NEP06.h5') as cdb:
>>> # find the available tags for existing db
>>> print(list(cdb.keys()))
['co', 'recombined', 'stack_86398', 'subdivision']
>>> # find available channels with tag subdivision
>>> print(cdb.get_available_channels(
>>> 'subdivision', 'XN-XN', 'NEP06-NEP06'))
['HHE-HHN', 'HHE-HHZ', 'HHN-HHZ']
>>> # Get Data from all times, specific channel and tag
>>> st = cdb.get_data(
>>> 'XN-XN', 'NEP06-NEP06', 'HHE-HHN', 'subdivision')
>>> print(st.count())
250
"""
if corr_options is None and mode != 'r' and not _force:
mode = 'r'
warnings.warn(
'Opening Correlation Databases without providing a correlation'
+ ' options dictionary is only allowed in read only mode.\n'
+ 'Setting mode read only `r`....')
# Create / read file
if not path.split('.')[-1] == 'h5':
path += '.h5'
self.path = path
self.mode = mode
self.compression = compression
self.co = corr_options
self.force = _force
def __enter__(self) -> DBHandler:
self.db_handler = DBHandler(
self.path, self.mode, self.compression, self.co, self.force)
return self.db_handler
def __exit__(self, exc_type, exc_value, tb) -> None or bool:
self.db_handler._close()
if exc_type is not None:
return False
[docs]def all_traces_recursive(
group: h5py._hl.group.Group, stream: CorrStream,
pattern: str) -> CorrStream:
"""
Recursively, appends all traces in a h5py group to the input stream.
In addition this will check whether the data matches a certain pattern.
:param group: group to search through
:type group: class:`h5py._hl.group.Group`
:param stream: Stream to append the traces to
:type stream: CorrStream
:param pattern: pattern for the path in the hdf5 file, see fnmatch for
details.
:type pattern: str
:return: Stream with appended traces
:rtype: CorrStream
"""
for v in group.values():
if isinstance(v, h5py._hl.group.Group):
all_traces_recursive(v, stream, pattern)
elif not fnmatch.fnmatch(v.name, pattern) and v.name not in pattern:
continue
else:
stream.append(
CorrTrace(np.array(v), _header=read_hdf5_header(v)))
return stream
[docs]def co_to_hdf5(co: dict) -> dict:
coc = deepcopy(co)
remk = [
'subdir', 'read_start', 'read_end', 'read_len', 'read_inc',
'combination_method', 'combinations', 'starttime',
'xcombinations', 'preprocess_subdiv', 'allow_different_params']
for key in remk:
coc.pop(key, None)
coc['corr_args'].pop('combinations', None)
coc['subdivision'].pop('recombine_subdivision', None)
coc['subdivision'].pop('delete_subdivision', None)
try:
[coc['preProcessing'].remove(step) for step in coc['preProcessing']
if 'stream_mask_at_utc' in step['function']]
except KeyError:
pass
return coc
