def __call__(self,
lparallel=False,
NP=None,
inner_list=None,
outer_list=None,
callback=None,
**kwargs):
''' this method is called instead of a class or instance method; it applies the arguments
'kwargs' to each ensemble member; it also supports argument expansion with inner and
outer product (prior to application to ensemble) and parallelization using multiprocessing '''
# expand kwargs to ensemble list
kwargs_list = expandArgumentList(inner_list=inner_list,
outer_list=outer_list,
**kwargs)
if len(kwargs_list) == 1:
kwargs_list = kwargs_list * len(self.klass.members)
elif len(kwargs_list) != len(self.klass.members):
raise ArgumentError(
'Length of expanded argument list does not match ensemble size! {} ~= {}'
.format(len(kwargs_list), len(self.klass.members)))
# loop over ensemble members and execute function
if lparallel:
# parallelize method execution using multiprocessing
pool = multiprocessing.Pool(processes=NP) # initialize worker pool
if callback is not None and not callable(callback):
raise TypeError(callback)
# N.B.: the callback function is passed a result from the apply_method function,
# which returns a tuple of the form (member, exit_code)
# define work loads (function and its arguments) and start tasks
results = [
pool.apply_async(apply_method, (member, self.attr),
kwargs,
callback=callback)
for member, kwargs in zip(self.klass.members, kwargs_list)
]
# N.B.: Beware Pickling!!!
pool.close()
pool.join() # wait to finish
# retrieve and assemble results
results = [result.get() for result in results]
# divide members and results (apply_method returns both, in case members were modified)
self.klass.members = [result[0] for result in results]
results = [result[1] for result in results]
else:
# get instance methods
methods = [
getattr(member, self.attr) for member in self.klass.members
]
# just apply sequentially
results = [
method(**kwargs)
for method, kwargs in zip(methods, kwargs_list)
]
if len(results) != len(self.klass.members):
raise ArgumentError(
'Length of results list does not match ensemble size! {} ~= {}'
.format(len(results), len(self.klass.members)))
return tuple(results)
def timeAxis(start_date=None,
end_date=None,
sampling=None,
date_range=None,
time_axis=None,
llastIncl=True,
ntime=None,
varatts=None):
''' figure out type and dimensions of time axis '''
# check time input
if date_range: start_date, end_date, sampling = date_range
if start_date and end_date and sampling:
start_year, start_month, start_day = convertDate(start_date)
start_datetime = np.datetime64(
dt.datetime(year=start_year, month=start_month, day=start_day),
sampling)
end_year, end_month, end_day = convertDate(end_date)
end_datetime = np.datetime64(
dt.datetime(year=end_year, month=end_month, day=end_day), sampling)
if llastIncl: end_datetime += np.timedelta64(1, sampling)
date_range = np.arange(start_datetime,
end_datetime,
dtype='datetime64[{}]'.format(sampling))
assert date_range[0] == start_datetime, date_range[0]
if ntime:
if ntime > len(date_range):
raise ArgumentError(date_range)
else:
# trim
date_range = date_range[0:ntime]
else:
ntime = len(date_range)
elif time_axis == 'datetime':
raise ArgumentError('Insufficient time axis information!')
# construct time axis
atts = varatts['time']
if time_axis.lower() == 'simple':
time = Axis(atts=atts, coord=np.arange(1, ntime + 1))
elif time_axis.lower() == 'datetime':
if sampling.lower() == 'y' or sampling.lower() == '1y': units = 'year'
elif sampling.lower() == 'm' or sampling.lower() == '1m':
units = 'month'
elif sampling.lower() == 'd' or sampling.lower() == '1d':
units = 'day'
elif sampling.lower() == 'h' or sampling.lower() == '1h':
units = 'hour'
else:
units = sampling
long_name = '{}s since {}'.format(units.title(), str(
date_range[0])) # hope this makes sense...
atts.update(long_name=long_name, units=units)
time = Axis(atts=atts, coord=date_range)
else:
raise ArgumentError(time_axis)
# return time axis
return time
def detrend(var,
ax=None,
lcopy=True,
ldetrend=True,
ltrend=False,
degree=1,
rcond=None,
w=None,
lsmooth=False,
lresidual=False,
window_len=11,
window='hanning'):
''' subtract a linear trend from a time-series array (operation is in-place) '''
# check input
if not isinstance(var, np.ndarray):
raise NotImplementedError # too many checks
if lcopy: var = var.copy() # make copy - not in-place!
# fit over entire array (usually not what we want...)
if ax is None and ldetrend:
ax = np.arange(var.size) # make dummy axis, if necessary
if var.ndim != 1:
shape = var.shape
var = var.ravel() # flatten array, if necessary
else:
shape = None
# apply optional detrending
if ldetrend or ltrend:
# fit linear trend
trend = np.polyfit(ax,
var,
deg=degree,
rcond=rcond,
w=w,
full=False,
cov=False)
# evaluate and subtract linear trend
if ldetrend and ltrend:
raise ArgumentError(
"Can either return trend/polyfit or residuals, not both.")
elif ldetrend and not ltrend:
var -= np.polyval(trend, ax) # residuals
elif ltrend and not ldetrend:
var = np.polyval(trend, ax) # residuals
# apply optional smoothing
if lsmooth and lresidual:
raise ArgumentError(
"Can either return smoothed array or residuals, not both.")
elif lsmooth:
var = smooth(var, window_len=window_len, window=window)
elif lresidual:
var -= smooth(var, window_len=window_len, window=window)
# return detrended and/or smoothed time-series
if shape is not None: var = var.reshape(shape)
return var
def getTimeseriesData(self, units='kg/s', lcheck=True, lexpand=True, lfill=True, period=None, lflatten=True):
''' extract time series data and time coordinates from a WSC monthly CSV file '''
if self.monthly_file:
# use numpy's CSV functionality
# get timeseries data
data = np.genfromtxt(self.monthly_file, dtype=np.float32, delimiter=',', skip_header=1, filling_values=np.nan,
usecols=np.arange(4,28,2), usemask=True, loose=True, invalid_raise=True)
assert data.shape[1] == 12, data.shape
# for some reason every value is followed by an extra comma...
#data = np.ma.masked_less(data, 10) # remove some invalid values
# N.B.: some values appear unrealistically small, however, these are removed in the check-
# section below (it appears they consistently fail the ckeck test)
if units.lower() == 'kg/s': data *= 1000. # m^3 == 1000 kg (water)
elif units.lower() == 'm^3/s': pass # original units
else: raise ArgumentError("Unknown units: {}".format(units))
# get time coordinates and verification flag
check = np.genfromtxt(self.monthly_file, dtype=np.int, delimiter=',', skip_header=1, filling_values=-9999,
usecols=np.arange(1,4,1), usemask=True, loose=True, invalid_raise=True)
assert check.shape[0] == data.shape[0], check.shape
assert np.all(check >= 0), np.sum(check < 0)
time = check[:,2].astype(np.int) # this is the year (time coordinate)
# determine valid entries
if lcheck:
check = np.all(check[:,:2]==1, axis=1) # require all entries to be one
# N.B.: I'm not sure what it means if values are not equal to one, but the flow values look
# unrealistically small (see above); probably different units...
data = data[check,:]; time = time[check]
assert time.shape[0] == data.shape[0], check.shape
# slice off values outside the period of interest
if period:
valid = np.logical_and(time >= period[0],time < period[1])
time = time[valid]; data = data[valid]
# fill in missing time periods/years
if lfill:
if period: time0 = period[0]; time1 = period[1]
else: time0 = time[0]; time1 = time[-1]+1
idx = np.asarray(time - time0, dtype=np.int32); tlen = time1 - time0 # start at 0; length is last value (+1)
pad_time = np.arange(time0,time1) # form continuous sequence
#assert np.all( pad_time[idx] == time ), idx # potentially expensive
time = pad_time # new continuous time coordinate
pad_data = np.ma.zeros((tlen,12), dtype=np.float32)*np.NaN # pre-allocate with NaN
pad_data.mask = True # mask everywhere for now
pad_data[idx,:] = data; #pad_data.mask[idx,:] = data.mask
#assert np.all( pad_data.mask[idx,:] == data.mask ) # potentially expensive
data = pad_data
# now, expand time coordinate by adding month
if lexpand:
time = time.reshape((time.size,1))
coord = np.repeat((time-1979)*12, 12, axis=1) + np.arange(0,12).reshape((1,12))
assert coord.shape == data.shape, coord.shape
#assert np.all( np.diff(coord.flatten()) == 1 ), coord # potentially expensive
time = coord
if lflatten:
time = time.flatten(); data = data.flatten()
# return data array and coordinate vector
return data, time
else:
raise IOError("No timeseries file defined or file not found for gage station '{}'.\n(folder: '{}')".format(self.name,self.folder))
def __init__(self, project=None, filetype='aux', folder=None, bc_method=None, **expargs):
''' take arguments that have been passed from caller and initialize parameters '''
if bc_method:
if not filetype: filetype = bc_method.lower()
elif filetype != bc_method.lower():
raise ArgumentError(filetype, bc_method)
self.bc_method = bc_method
self.filetype = filetype; self.folder_pattern = folder
self.export_arguments = expargs
def binedges(bins=None, binedgs=None, limits=None, lcheckVar=True):
''' utility function to generate and validate bins and binegdes from either one '''
# check input
if bins is None and binedgs is None: raise ArgumentError
elif bins is not None and binedgs is not None:
if len(bins) + 1 != len(binedgs): raise ArgumentError(len(bins))
if bins is not None:
if limits is not None: vmin, vmax = limits
else: raise ArgumentError(bins)
# expand bins (values refer to center of bins)
if isinstance(bins, (int, np.integer)):
if bins == 1: bins = np.asarray(((vmin + vmax) / 2., ))
else: bins = np.linspace(vmin, vmax, bins)
elif isinstance(bins, (tuple, list)) and 0 < len(bins) < 4:
bins = np.linspace(*bins)
elif not isinstance(bins, (list, np.ndarray)):
raise TypeError(bins)
if len(bins) == 1:
tmpbinedgs = np.asarray((vmin, vmax))
else:
hbd = np.diff(bins) / 2. # make sure this is a float!
tmpbinedgs = np.hstack(
(bins[0] - hbd[0], bins[1:] - hbd,
bins[-1] + hbd[-1])) # assuming even spacing
if binedgs is None: binedgs = tmpbinedgs # computed from bins
elif lcheckVar:
assert isEqual(binedgs, np.asarray(tmpbinedgs,
dtype=binedgs.dtype))
if binedgs is not None:
# expand bin edges
if not isinstance(binedgs, (tuple, list)):
binedgs = np.asarray(binedgs)
elif not isinstance(binedgs, np.ndarray):
raise TypeError(binedgs)
tmpbins = binedgs[1:] - (np.diff(binedgs) / 2.
) # make sure this is a float!
if bins is None: bins = tmpbins # compute from binedgs
elif lcheckVar:
assert isEqual(bins, np.asarray(tmpbins, dtype=bins.dtype))
# return bins and binegdes
return bins, binedgs
def loadMetadata(well,
filename='metadata.dbf',
wellname='W{WELL_ID:07d}-{WELL_NO:1d}',
llistWells=False,
folder=None,
conservation_authority=None):
if not folder and conservation_authority:
folder = ca_folder.format(conservation_authority)
# clean up well name
well_id, well_no = getWellName(well)
well = wellname.format(WELL_ID=well_id, WELL_NO=well_no)
# open database and get relevant entry
#from simpledbf import Dbf5
from dbfread import DBF
filepath = filename if folder is None else os.path.join(folder, filename)
table = DBF(filepath)
meta = None
for record in table:
if llistWells: print((record['PGMN_WELL']))
if record['PGMN_WELL'] == well:
meta = record.copy()
if meta is None:
raise ArgumentError(well)
# parse screen information
screen_type, screen_depth = meta['SCREEN_HOL'].split(':')
meta['Screen'] = screen_type.title()
screen_hilo = []
lunit = False
for hilo in screen_depth.split('-'):
if hilo[-1] == 'M':
lunit = True
screen_hilo.append(float(hilo[:-1]))
else:
screen_hilo.append(float(hilo))
if not lunit: raise ValueError(screen_depth)
assert len(screen_hilo) == 2, screen_hilo
meta['screen_top'] = screen_hilo[0]
meta['screen_bottom'] = screen_hilo[1]
meta['screen_depth'] = (screen_hilo[0] + screen_hilo[1]) / 2.
meta['zs'] = float(meta['ELVA_GROUN'])
meta['z'] = meta['zs'] - meta['screen_depth']
meta['z_t'] = meta['zs'] - meta['screen_top']
meta['z_b'] = meta['zs'] - meta['screen_bottom']
# return coordinate arrays (in degree)
return meta
def __init__(self, basin=None, river=None, name=None, folder=None, lcheck=False):
''' initialize gage station based on various input data '''
if name is None: raise ArgumentError()
if folder is None: folder = '{:s}/Basins/{:s}/'.format(root_folder,basin)
if not os.path.isdir(folder): IOError(folder)
if river is not None and river not in name: name = '{:s}_{:s}'.format(river,name)
if not os.path.isdir(folder): IOError(folder)
self.folder = folder # usually basin folder
self.name = name # or prefix...
self.basin_name = basin # has to be a long_name in order to construct the folder
self.meta_file = '{:s}/{:s}'.format(folder,name + self.meta_ext)
if not os.path.isfile(self.meta_file):
if lcheck: raise IOError(self.meta_file)
else: self.meta_file = None # clear if not available
self.monthly_file = '{:s}/{:s}'.format(folder,name + self.monthly_ext)
if not os.path.isfile(self.monthly_file):
if lcheck: raise IOError(self.monthly_file)
else: self.monthly_file = None # clear if not available
def loadHGS_StnEns(ensemble=None, station=None, varlist=None, varatts=None, name=None, title=None,
period=None, run_period=15, folder=None, obs_period=None,
ensemble_list=None, ensemble_args=None, observation_list=None, # ensemble and obs lists for project
loadHGS_StnTS=loadHGS_StnTS, loadWSC_StnTS=loadWSC_StnTS, # these can also be overloaded
prefix=None, WSC_station=None, basin=None, basin_list=None, **kwargs):
''' a wrapper for the regular HGS loader that can also load gage stations and assemble ensembles '''
if observation_list is None: observation_list = ('obs','observations')
if ensemble_list is None: ensemble_list = dict() # empty, i.e. no ensembles
elif not isinstance(ensemble_list, dict): raise TypeError(ensemble_list)
if ensemble is None: raise ArgumentError("Mandatory argument 'ensemble' is not defined!")
# decide what to do, based on inputs
if ensemble.lower() in observation_list:
# translate parameters
station = station if WSC_station is None else WSC_station
period = period if obs_period is None else obs_period
filetype = 'monthly'
# load gage station with slightly altered parameters
dataset = loadWSC_StnTS(station=station, name=name, title=title, basin=basin, basin_list=basin_list,
varlist=varlist, varatts=varatts, period=period, filetype=filetype)
elif ensemble.lower() in ensemble_list:
if ensemble_args is None: ensemble_args = dict()
# loop over list of experiments in ensemble
ens = []
for exp in ensemble_list[ensemble]:
# load individual HGS simulation
ds = loadHGS_StnTS(station=station, varlist=varlist, varatts=varatts, name=name, title=title,
period=period, ENSEMBLE=exp, run_period=run_period, folder=folder, prefix=prefix,
WSC_station=WSC_station, basin=basin, basin_list=basin_list, **kwargs)
ens.append(ds)
# construct ensemble by concatenating time-series
ensemble_args.setdefault('name',ds.name.replace(exp,ensemble).replace(exp.title(),ensemble.title()))
ensemble_args.setdefault('title',ds.title.replace(exp,ensemble).replace(exp.title(),ensemble.title()))
# N.B.: the ensemble name is constructed by replacing the experiment name in specific dataset names with the ensemble name
ensemble_args.setdefault('axis','time')
dataset = concatDatasets(ens, **ensemble_args)
else:
# load HGS simulation
dataset = loadHGS_StnTS(station=station, varlist=varlist, varatts=varatts, name=name, title=title, period=period,
ENSEMBLE=ensemble, run_period=run_period, folder=folder, prefix=prefix,
WSC_station=WSC_station, basin=basin, basin_list=basin_list, **kwargs)
return dataset
def PCA(data,
degree=None,
lprewhiten=False,
lpostwhiten=False,
lEOF=False,
lfeedback=False):
''' A function to perform principal component analysis and return the time-series of the leading EOF's. '''
data = np.asarray(data)
if not data.ndim == 2: raise ArgumentError(data.ndim)
# pre-whiten features
if lprewhiten:
data -= data.mean(axis=0, keepdims=True)
data /= data.std(axis=0, keepdims=True)
# compute PCA
R = np.cov(data.transpose()) # covariance matrix
eig, eof = la.eigh(R) # eigenvalues, eigenvectors (of symmetric matrix)
ieig = np.argsort(eig, )[::-1] # sort in descending order
eig = eig[ieig]
eof = eof[:, ieig]
eig /= eig.sum() # normalize by total variance
# truncate EOF's
if degree is not None:
eig = eig[:degree]
eof = eof[:, :degree]
# generate report/feedback
if lfeedback:
string = "Variance explained by {:s} PCA's: {:s}; total variance explained: {:2.0f}%"
eiglist = ', '.join('{:.0f}%'.format(e * 100.) for e in eig)
dgrstr = 'all' if degree is None else "{:d} leading".format(degree)
print(string.format(dgrstr, eiglist, eig.sum() * 100.))
# project data onto (leading) EOF's
pca = np.dot(data, eof) # inverse order, because the are transposed
# post-whiten features
if lpostwhiten:
pca -= pca.mean(axis=0, keepdims=True)
pca /= pca.std(axis=0, keepdims=True)
# return results
if lEOF: return pca, eig, eof
else: return pca, eig
def loadHydro(filename='discharge_out.mpiio', folder=None, nreal=None, ntime=None, dtype=np.float64):
''' function to load hydroggraphs/discharge from EnKF output '''
if not nreal and not ntime:
raise ArgumentError("Please specify number of realizations 'nreal' or number of time steps 'ntime'.")
filepath = os.path.join(folder,filename)
if isinstance(dtype, str): dtype = getattr(np,dtype)
# load data
data = np.fromfile(filepath, dtype=dtype)
# reshape (need to know number or realizations or time steps)
n = data.size
if nreal:
nt = int(n/nreal)
nr = nreal
if ntime and nt != ntime:
raise ValueError("Given number of time steps is not consistent with file size or data type ({} != {}).".format(ntime,nt))
elif ntime:
nt = ntime
nr = int(n/ntime)
if nt*nr != n:
raise ValueError("Given number of realizations ({}) and time steps do not divide number of data points ({}).".format(nr,nt,n))
data = data.reshape((nt,nr))
# return timeseries
return data
def defaultNamedtuple(typename, field_names, defaults=None):
''' wrapper for namedtuple that supports defaults; adapted from stackoverflow:
https://stackoverflow.com/questions/11351032/named-tuple-and-optional-keyword-arguments '''
T = col.namedtuple(typename, field_names) # make named tuple
T.__new__.__defaults__ = (None, ) * len(T._fields) # set defaults to None
# add custom defaults
if defaults is not None:
if isinstance(defaults, col.Mapping):
prototype = T(**defaults)
elif isinstance(defaults, col.Iterable):
prototype = T(*defaults)
else:
raise ArgumentError(str(defaults))
T.__new__.__defaults__ = tuple(prototype)
# # add self-referenc defaults
# if ref_prefix:
# l = len(ref_prefix)
# for field,value in T._asdict().iteritems():
# if isinstance(value,basestring) and value[:l] == ref_prefix:
# T.__dict__[field] = T.__dict__[value[l:]]
# # N.B.: this would have to go into the constructor in order to work...
# return namedtuple with defaults
return T
def loadHGS_StnTS(station=None, varlist=None, varatts=None, folder=None, name=None, title=None,
start_date=None, end_date=None, run_period=15, period=None, lskipNaN=False, lcheckComplete=True,
basin=None, WSC_station=None, basin_list=None, filename=None, prefix=None,
scalefactors=None, **kwargs):
''' Get a properly formatted WRF dataset with monthly time-series at station locations; as in
the hgsrun module, the capitalized kwargs can be used to construct folders and/or names '''
if folder is None or ( filename is None and station is None ): raise ArgumentError
# try to find meta data for gage station from WSC
HGS_station = station
if basin is not None and basin_list is not None:
station_name = station
station = getGageStation(basin=basin, station=station if WSC_station is None else WSC_station,
basin_list=basin_list) # only works with registered basins
if station_name is None: station_name = station.name # backup, in case we don't have a HGS station name
metadata = station.getMetaData() # load station meta data
if metadata is None: raise GageStationError(name)
else:
metadata = dict(); station = None; station_name = None
# prepare name expansion arguments (all capitalized)
expargs = dict(ROOT_FOLDER=root_folder, STATION=HGS_station, NAME=name, TITLE=title,
PREFIX=prefix, BASIN=basin, WSC_STATION=WSC_station)
for key,value in metadata.items():
if isinstance(value,basestring):
expargs['WSC_'+key.upper()] = value # in particular, this includes WSC_ID
if 'WSC_ID' in expargs:
if expargs['WSC_ID'][0] == '0': expargs['WSC_ID0'] = expargs['WSC_ID'][1:]
else: raise DatasetError('Expected leading zero in WSC station ID: {}'.format(expargs['WSC_ID']))
# exparg preset keys will get overwritten if capitalized versions are defined
for key,value in kwargs.items():
KEY = key.upper() # we only use capitalized keywords, and non-capitalized keywords are only used/converted
if KEY == key or KEY not in kwargs: expargs[KEY] = value # if no capitalized version is defined
# read folder and infer prefix, if necessary
folder = folder.format(**expargs)
if not os.path.exists(folder): raise IOError(folder)
if expargs['PREFIX'] is None:
with open('{}/{}'.format(folder,prefix_file), 'r') as pfx:
expargs['PREFIX'] = prefix = ''.join(pfx.readlines()).strip()
# now assemble file name for station timeseries
filename = filename.format(**expargs)
filepath = '{}/{}'.format(folder,filename)
if not os.path.exists(filepath): IOError(filepath)
if station_name is None:
station_name = filename[filename.index('hydrograph.')+1:-4] if station is None else station
# set meta data (and allow keyword expansion of name and title)
metadata['problem'] = prefix
metadata['station_name'] = metadata.get('long_name', station_name)
if name is not None: name = name.format(**expargs) # name expansion with capitalized keyword arguments
else: name = 'HGS_{:s}'.format(station_name)
metadata['name'] = name; expargs['Name'] = name.title() # name in title format
if title is None: title = '{{Name:s}} (HGS, {problem:s})'.format(**metadata)
title = title.format(**expargs) # name expansion with capitalized keyword arguments
metadata['long_name'] = metadata['title'] = title
# now determine start data for date_parser
if end_date is None:
if start_date and run_period: end_date = start_date + run_period
elif period: end_date = period[1]
else: raise ArgumentError("Need to specify either 'start_date' & 'run_period' or 'period' to infer 'end_date'.")
end_year,end_month,end_day = convertDate(end_date)
if start_date is None:
if end_date and run_period: start_date = end_date - run_period
elif period: start_date = period[0]
else: raise ArgumentError("Need to specify either 'end_date' & 'run_period' or 'period' to infer 'start_date'.")
start_year,start_month,start_day = convertDate(start_date)
if start_day != 1 or end_day != 1:
raise NotImplementedError('Currently only monthly data is supported.')
# import functools
# date_parser = functools.partial(date_parser, year=start_year, month=start_month, day=start_day)
# # now load data using pandas ascii reader
# data_frame = pd.read_table(filepath, sep='\s+', header=2, dtype=np.float64, index_col=['time'],
# date_parser=date_parser, names=ascii_varlist)
# # resample to monthly data
# data_frame = data_frame.resample(resampling).agg(np.mean)
# data = data_frame[flowvar].values
# parse header
if varlist is None: varlist = variable_list[:] # default list
with open(filepath, 'r') as f:
line = f.readline(); lline = line.lower() # 1st line
if not "hydrograph" in lline: raise GageStationError(line,filepath)
# parse variables and determine columns
line = f.readline(); lline = line.lower() # 2nd line
if not "variables" in lline: raise GageStationError(line)
variable_order = [v.strip('"').lower() for v in line[line.find('"'):].strip().split(',')]
# figure out varlist and data columns
if variable_order[0] == 'time': del variable_order[0] # only keep variables
else: raise GageStationError(variable_order)
variable_order = [hgs_variables[v] for v in variable_order] # replace HGS names with GeoPy names
vardict = {v:i+1 for i,v in enumerate(variable_order)} # column mapping; +1 because time was removed
variable_order = [v for v in variable_order if v in varlist or flow_to_flux[v] in varlist]
usecols = tuple(vardict[v] for v in variable_order) # variable columns that need to loaded (except time, which is col 0)
assert 0 not in usecols, usecols
# load data as tab separated values
data = np.genfromtxt(filepath, dtype=np.float64, delimiter=None, skip_header=3, usecols = (0,)+usecols)
assert data.shape[1] == len(usecols)+1, data.shape
if lskipNaN:
data = data[np.isnan(data).sum(axis=1)==0,:]
elif np.any( np.isnan(data) ):
raise DataError("Missing values (NaN) encountered in hydrograph file; use 'lskipNaN' to ignore.\n('{:s}')".format(filepath))
time_series = data[:,0]; flow_data = data[:,1:]
assert flow_data.shape == (len(time_series),len(usecols)), flow_data.shape
# original time deltas in seconds
time_diff = time_series.copy(); time_diff[1:] = np.diff(time_series) # time period between time steps
assert np.all( time_diff > 0 ), filepath
time_diff = time_diff.reshape((len(time_diff),1)) # reshape to make sure broadcasting works
# integrate flow over time steps before resampling
flow_data[1:,:] -= np.diff(flow_data, axis=0)/2. # get average flow between time steps
flow_data *= time_diff # integrate flow in time interval by multiplying average flow with time period
flow_data = np.cumsum(flow_data, axis=0) # integrate by summing up total flow per time interval
# generate regular monthly time steps
start_datetime = np.datetime64(dt.datetime(year=start_year, month=start_month, day=start_day), 'M')
end_datetime = np.datetime64(dt.datetime(year=end_year, month=end_month, day=end_day), 'M')
time_monthly = np.arange(start_datetime, end_datetime+np.timedelta64(1, 'M'), dtype='datetime64[M]')
assert time_monthly[0] == start_datetime, time_monthly[0]
assert time_monthly[-1] == end_datetime, time_monthly[-1]
# convert monthly time series to regular array of seconds since start date
time_monthly = ( time_monthly.astype('datetime64[s]') - start_datetime.astype('datetime64[s]') ) / np.timedelta64(1,'s')
assert time_monthly[0] == 0, time_monthly[0]
# interpolate integrated flow to new time axis
#flow_data = np.interp(time_monthly, xp=time_series[:,0], fp=flow_data[:,0],).reshape((len(time_monthly),1))
time_series = np.concatenate(([0],time_series), axis=0) # integrated flow at time zero must be zero...
flow_data = np.concatenate(([[0,]*len(usecols)],flow_data), axis=0) # ... this is probably better than interpolation
# N.B.: we are adding zeros here so we don't have to extrapolate to the left; on the right we just fill in NaN's
if ( time_monthly[-1] - time_series[-1] ) > 3*86400. and lcheckComplete:
warn("Data record ends more than 3 days befor end of period: {} days".format((time_monthly[-1]-time_series[-1])/86400.))
elif (time_monthly[-1]-time_series[-1]) > 5*86400.:
if lcheckComplete:
raise DataError("Data record ends more than 5 days befor end of period: {} days".format((time_monthly[-1]-time_series[-1])/86400.))
else:
warn("Data record ends more than 5 days befor end of period: {} days".format((time_monthly[-1]-time_series[-1])/86400.))
flow_interp = si.interp1d(x=time_series, y=flow_data, kind='linear', axis=0, copy=False,
bounds_error=False, fill_value=np.NaN, assume_sorted=True)
flow_data = flow_interp(time_monthly) # evaluate with call
# compute monthly flow rate from interpolated integrated flow
flow_data = np.diff(flow_data, axis=0) / np.diff(time_monthly, axis=0).reshape((len(time_monthly)-1,1))
flow_data *= 1000 # convert from m^3/s to kg/s
# construct time axis
start_time = 12*(start_year - 1979) + start_month -1
end_time = 12*(end_year - 1979) + end_month -1
time = Axis(name='time', units='month', atts=dict(long_name='Month since 1979-01'),
coord=np.arange(start_time, end_time)) # not including the last, e.g. 1979-01 to 1980-01 is 12 month
assert len(time_monthly) == end_time-start_time+1
assert flow_data.shape == (len(time),len(variable_order)), (flow_data.shape,len(time),len(variable_order))
# construct dataset
dataset = Dataset(atts=metadata)
dataset.station = station # add gage station object, if available (else None)
for i,flowvar in enumerate(variable_order):
data = flow_data[:,i]
fluxvar = flow_to_flux[flowvar]
if flowvar in varlist:
flowatts = variable_attributes[flowvar]
# convert variables and put into dataset (monthly time series)
if flowatts['units'] != 'kg/s':
raise VariableError("Hydrograph data is read as kg/s; flow variable does not match.\n{}".format(flowatts))
dataset += Variable(data=data, axes=(time,), **flowatts)
if fluxvar in varlist and 'shp_area' in metadata:
# compute surface flux variable based on drainage area
fluxatts = variable_attributes[fluxvar]
if fluxatts['units'] == 'kg/s' and fluxatts['units'] != 'kg/m^2/s': raise VariableError(fluxatts)
data = data / metadata['shp_area'] # need to make a copy
dataset += Variable(data=data, axes=(time,), **fluxatts)
# apply analysis period
if period is not None:
dataset = dataset(years=period)
# adjust scalefactors, if necessary
if scalefactors:
if isinstance(scalefactors,dict):
dataset = updateScalefactor(dataset, varlist=scalefactors, scalefactor=None)
elif isNumber(scalefactors):
scalelist = ('discharge','seepage','flow')
dataset = updateScalefactor(dataset, varlist=scalelist, scalefactor=scalefactors)
else:
raise TypeError(scalefactors)
# return completed dataset
return dataset
def loadGageStation(basin=None, station=None, varlist=None, varatts=None, mode='climatology',
aggregation=None, filetype='monthly', folder=None, name=None, period=None,
basin_list=None, lcheck=True, lexpand=True, lfill=True, lflatten=True,
lkgs=True, scalefactors=None, title=None):
''' function to load hydrograph climatologies and timeseries for a given basin '''
## resolve input
if mode == 'timeseries' and aggregation:
raise ArgumentError('Timeseries does not support aggregation.')
# get GageStation instance
station = getGageStation(basin=basin, station=station, name=name, folder=folder,
river=None, basin_list=basin_list, lcheck=True)
# variable attributes
if varlist is None: varlist = variable_list
elif not isinstance(varlist,(list,tuple)): raise TypeError
varlist = list(varlist) # make copy of varlist to avoid interference
if varatts is None:
if aggregation is None: varatts = variable_attributes_kgs if lkgs else variable_attributes_mms
else: varatts = agg_varatts_kgs if lkgs else agg_varatts_mms
elif not isinstance(varatts,dict): raise TypeError
## read csv data
# time series data and time coordinates
lexpand = True; lfill = True
if mode == 'climatology': lexpand = False; lfill = False; lflatten = False
data, time = station.getTimeseriesData(units='kg/s' if lkgs else 'm^3/s', lcheck=True, lexpand=lexpand,
lfill=lfill, period=period, lflatten=lflatten)
# station meta data
metadata = station.getMetaData(lcheck=True)
den = metadata['shp_area'] if lkgs else ( metadata['shp_area'] / 1000. )
## create dataset for station
dataset = Dataset(name='WSC', title=title or metadata['Station Name'], varlist=[], atts=metadata,)
if mode.lower() in ('timeseries','time-series'):
time = time.flatten(); data = data.flatten() # just to make sure...
# make time axis based on time coordinate from csv file
timeAxis = Axis(name='time', units='month', coord=time, # time series centered at 1979-01
atts=dict(long_name='Month since 1979-01'))
dataset += timeAxis
# load mean discharge
dataset += Variable(axes=[timeAxis], data=data, atts=varatts['discharge'])
# load mean runoff
doa = data / den
dataset += Variable(axes=[timeAxis], data=doa, atts=varatts['runoff'])
elif mode == 'climatology':
# N.B.: this is primarily for backwards compatibility; it should not be used anymore...
# make common time axis for climatology
te = 12 # length of time axis: 12 month
climAxis = Axis(name='time', units='month', length=12, coord=np.arange(1,te+1,1)) # monthly climatology
dataset.addAxis(climAxis, copy=False)
# extract variables (min/max/mean are separate variables)
# N.B.: this is mainly for backwards compatibility
doa = data / den
if aggregation is None or aggregation.lower() == 'mean':
# load mean discharge
tmpdata = nf.nanmean(data, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['discharge'])
dataset.addVariable(tmpvar, copy=False)
# load mean runoff
tmpdata = nf.nanmean(doa, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['runoff'])
dataset.addVariable(tmpvar, copy=False)
if aggregation is None or aggregation.lower() == 'std':
# load discharge standard deviation
tmpdata = nf.nanstd(data, axis=0, ddof=1) # very few values means large uncertainty!
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['discstd'])
dataset.addVariable(tmpvar, copy=False)
# load runoff standard deviation
tmpdata = nf.nanstd(doa, axis=0, ddof=1)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['roff_std'])
dataset.addVariable(tmpvar, copy=False)
if aggregation is None or aggregation.lower() == 'sem':
# load discharge standard deviation
tmpdata = nf.nansem(data, axis=0, ddof=1) # very few values means large uncertainty!
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['discsem'])
dataset.addVariable(tmpvar, copy=False)
# load runoff standard deviation
tmpdata = nf.nansem(doa, axis=0, ddof=1)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['roff_sem'])
dataset.addVariable(tmpvar, copy=False)
if aggregation is None or aggregation.lower() == 'max':
# load maximum discharge
tmpdata = nf.nanmax(data, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['discmax'])
dataset.addVariable(tmpvar, copy=False)
# load maximum runoff
tmpdata = nf.nanmax(doa, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['roff_max'])
dataset.addVariable(tmpvar, copy=False)
if aggregation is None or aggregation.lower() == 'min':
# load minimum discharge
tmpdata = nf.nanmin(data, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['discmin'])
dataset.addVariable(tmpvar, copy=False)
# load minimum runoff
tmpdata = nf.nanmin(doa, axis=0)
tmpvar = Variable(axes=[climAxis], data=tmpdata, atts=varatts['roff_min'])
dataset.addVariable(tmpvar, copy=False)
else:
raise NotImplementedError, "Time axis mode '{}' is not supported.".format(mode)
# adjust scalefactors, if necessary
if scalefactors:
if isinstance(scalefactors,dict):
dataset = updateScalefactor(dataset, varlist=scalefactors, scalefactor=None)
elif isNumber(scalefactors):
scalelist = ('discharge','StdDisc','SEMDisc','MaxDisc','MinDisc',)
dataset = updateScalefactor(dataset, varlist=scalelist, scalefactor=scalefactors)
else:
raise TypeError(scalefactors)
# return station dataset
return dataset
def rasterVariable(name=None,
units=None,
axes=None,
atts=None,
plot=None,
dtype=None,
projection=None,
griddef=None,
file_pattern=None,
lgzip=None,
lgdal=True,
lmask=True,
fillValue=None,
lskipMissing=True,
path_params=None,
offset=0,
scalefactor=1,
transform=None,
time_axis=None,
lfeedback=False,
**kwargs):
''' function to read multi-dimensional raster data and construct a GDAL-enabled Variable object '''
# print status
if lfeedback: print "Loading variable '{}': ".format(name), # no newline
## figure out axes arguments and load data
# figure out axes (list/tuple of axes has to be ordered correctly!)
axes_list = [ax.name for ax in axes[:-2]]
# N.B.: the last two axes are the two horizontal map axes (x&y); they can be None and will be inferred from raster
# N.B.: coordinate values can be overridden with keyword arguments, but length must be consistent
# figure out coordinates for axes
for ax in axes[:-2]:
if ax.name in kwargs:
# just make sure the dimensions match, but use keyword argument
if not len(kwargs[ax.name]) == len(ax):
raise AxisError(
"Length of Variable axis and raster file dimension have to be equal."
)
else:
# use Axis coordinates and add to kwargs for readRasterArray call
kwargs[ax.name] = tuple(ax.coord)
# load raster data
if lfeedback: print("'{}'".format(file_pattern))
data, geotransform = readRasterArray(file_pattern,
lgzip=lgzip,
lgdal=lgdal,
dtype=dtype,
lmask=lmask,
fillValue=fillValue,
lgeotransform=True,
axes=axes_list,
lna=False,
lskipMissing=lskipMissing,
path_params=path_params,
lfeedback=lfeedback,
**kwargs)
# shift and rescale
if offset != 0: data += offset
if scalefactor != 1: data *= scalefactor
## create Variable object and add GDAL
# check map axes and generate if necessary
xlon, ylat = getAxes(
geotransform,
xlen=data.shape[-1],
ylen=data.shape[-2],
projected=griddef.isProjected if griddef else bool(projection))
axes = list(axes)
if axes[-1] is None: axes[-1] = xlon
elif len(axes[-1]) != len(xlon): raise AxisError(axes[-1])
if axes[-2] is None: axes[-2] = ylat
elif len(axes[-2]) != len(ylat): raise AxisError(axes[-2])
# create regular Variable with data in memory
var = Variable(name=name,
units=units,
axes=axes,
data=data,
dtype=dtype,
mask=None,
fillValue=fillValue,
atts=atts,
plot=plot)
# apply transform (if any), now that we have axes etc.
if transform is not None: var = transform(var=var, time_axis=time_axis)
# add GDAL functionality
if griddef is not None:
# perform some consistency checks ...
if projection is None:
projection = griddef.projection
elif projection != griddef.projection:
raise ArgumentError(
"Conflicting projection and GridDef!\n {} != {}".format(
projection, griddef.projection))
if not np.isclose(geotransform, griddef.geotransform).all():
raise ArgumentError(
"Conflicting geotransform (from raster) and GridDef!\n {} != {}"
.format(geotransform, griddef.geotransform))
# ... and use provided geotransform (due to issues with numerical precision, this is usually better)
geotransform = griddef.geotransform # if we don't pass the geotransform explicitly, it will be recomputed from the axes
# add GDAL functionality
var = addGDALtoVar(var,
griddef=griddef,
projection=projection,
geotransform=geotransform,
gridfolder=None)
# return final, GDAL-enabled variable
return var
def rasterDataset(name=None,
title=None,
vardefs=None,
axdefs=None,
atts=None,
projection=None,
griddef=None,
lgzip=None,
lgdal=True,
lmask=True,
fillValue=None,
lskipMissing=True,
lgeolocator=True,
file_pattern=None,
lfeedback=True,
**kwargs):
''' function to load a set of variables that are stored in raster format in a systematic directory tree into a Dataset
Variables and Axis are defined as follows:
vardefs[varname] = dict(name=string, units=string, axes=tuple of strings, atts=dict, plot=dict, dtype=np.dtype, fillValue=value)
axdefs[axname] = dict(name=string, units=string, atts=dict, coord=array or list) or None
The path to raster files is constructed as variable_pattern+axes_pattern, where axes_pattern is defined through the axes,
(as in rasterVarialbe) and variable_pattern takes the special keywords VAR, which is the variable key in vardefs.
'''
## prepare input data and axes
if griddef:
xlon, ylat = griddef.xlon, griddef.ylat
if projection is None:
projection = griddef.projection
elif projection != griddef.projection:
raise ArgumentError("Conflicting projection and GridDef!")
geotransform = griddef.geotransform
isProjected = griddef.isProjected
else:
xlon = ylat = geotransform = None
isProjected = False if projection is None else True
# construct axes dict
axes = dict()
for axname, axdef in axdefs.items():
assert 'coord' in axdef, axdef
assert ('name' in axdef and 'units' in axdef) or 'atts' in axdef, axdef
if axdef is None:
axes[axname] = None
else:
ax = Axis(**axdef)
axes[ax.name] = ax
# check for map Axis
if isProjected:
if 'x' not in axes: axes['x'] = xlon
if 'y' not in axes: axes['y'] = ylat
else:
if 'lon' not in axes: axes['lon'] = xlon
if 'lat' not in axes: axes['lat'] = ylat
## load raster data into Variable objects
varlist = []
for varname, vardef in vardefs.items():
# check definitions
assert 'axes' in vardef and 'dtype' in vardef, vardef
assert ('name' in vardef
and 'units' in vardef) or 'atts' in vardef, vardef
# determine relevant axes
vardef = vardef.copy()
axes_list = [
None if ax is None else axes[ax] for ax in vardef.pop('axes')
]
# define path parameters (with varname)
path_params = vardef.pop('path_params', None)
path_params = dict() if path_params is None else path_params.copy()
if 'VAR' not in path_params:
path_params['VAR'] = varname # a special key
# add kwargs and relevant axis indices
relaxes = [ax.name for ax in axes_list
if ax is not None] # relevant axes
for key, value in kwargs.items():
if key not in axes or key in relaxes:
vardef[key] = value
# create Variable object
var = rasterVariable(projection=projection,
griddef=griddef,
file_pattern=file_pattern,
lgzip=lgzip,
lgdal=lgdal,
lmask=lmask,
lskipMissing=lskipMissing,
axes=axes_list,
path_params=path_params,
lfeedback=lfeedback,
**vardef)
# vardef components: name, units, atts, plot, dtype, fillValue
varlist.append(var)
# check that map axes are correct
for ax in var.xlon, var.ylat:
if axes[ax.name] is None: axes[ax.name] = ax
elif axes[ax.name] != ax:
raise AxisError("{} axes are incompatible.".format(ax.name))
if griddef is None: griddef = var.griddef
elif griddef != var.griddef:
raise AxisError("GridDefs are inconsistent.")
if geotransform is None: geotransform = var.geotransform
elif geotransform != var.geotransform:
raise AxisError(
"Conflicting geotransform (from Variable) and GridDef!\n {} != {}"
.format(var.geotransform, geotransform))
## create Dataset
# create dataset
dataset = Dataset(name=name,
title=title,
varlist=varlist,
axes=axes,
atts=atts)
# add GDAL functionality
dataset = addGDALtoDataset(dataset,
griddef=griddef,
projection=projection,
geotransform=geotransform,
gridfolder=None,
lwrap360=None,
geolocator=lgeolocator,
lforce=False)
# N.B.: for some reason we also need to pass the geotransform, otherwise it is recomputed internally and some consistency
# checks fail due to machine-precision differences
# return GDAL-enabled Dataset
return dataset
def expandArgumentList(inner_list=None,
outer_list=None,
expand_list=None,
lproduct='outer',
**kwargs):
''' A function that generates a list of complete argument dict's, based on given kwargs and certain
expansion rules: kwargs listed in expand_list are expanded and distributed element-wise,
either as inner ('inner_list') or outer ('outer_list') product, while other kwargs are repeated
in every argument dict.
Arguments can be expanded simultaneously (in parallel) within an outer product by specifying
them as a tuple within the outer product argument list ('outer_list'). '''
if not (expand_list or inner_list or outer_list):
arg_dicts = [kwargs] # return immediately - nothing to do
else:
# handle legacy arguments
if expand_list is not None:
if inner_list is not None or outer_list is not None:
raise ArgumentError("Can not mix input modes!")
if lproduct.lower() == 'inner': inner_list = expand_list
elif lproduct.lower() == 'outer': outer_list = expand_list
else: raise ArgumentError(lproduct)
outer_list = outer_list or []
inner_list = inner_list or []
# handle outer product expansion first
if len(outer_list) > 0:
kwtmp = {
key: value
for key, value in kwargs.items() if key not in inner_list
}
# detect variables for parallel expansion
# N.B.: parallel outer expansion is handled by replacing the arguments in each parallel expansion group
# with a single (fake) argument that is a tuple of the original argument values; the tuple is then,
# after expansion, disassembled into its former constituent arguments
par_dict = dict()
for kw in outer_list:
if isinstance(kw, (tuple, list)):
# retrieve parallel expansion group
par_args = [kwtmp.pop(name) for name in kw]
if not all(
[len(args) == len(par_args[0]) for args in par_args]):
raise ArgumentError(
"Lists for parallel expansion arguments have to be of same length!"
)
# introduce fake argument and save record
fake = 'TMP_' + '_'.join(kw) + '_{:d}'.format(
len(kw)) # long name that is unlikely to interfere...
par_dict[
fake] = kw # store record of parallel expansion for reassembly later
kwtmp[fake] = zip(
*par_args) # transpose lists to get a list of tuples
elif not isinstance(kw, basestring):
raise TypeError(kw)
# replace entries in outer list
if len(par_dict) > 0:
outer_list = outer_list[:] # copy list
for fake, names in par_dict.items():
if names in outer_list:
outer_list[outer_list.index(names)] = fake
assert all([isinstance(arg, basestring) for arg in outer_list])
outer_list, outer_dict = _prepareList(outer_list, kwtmp)
lstlen = 1
for el in outer_list:
lstlen *= len(outer_dict[el])
# execute recursive function for outer product expansion
list_dict = _loop_recursion(outer_list,
**outer_dict) # use copy of
# N.B.: returns a dictionary where all kwargs have been expanded to lists of appropriate length
assert all(key in outer_dict for key in list_dict.iterkeys())
assert all(len(list_dict[el]) == lstlen
for el in outer_list) # check length
assert all(len(ld) == lstlen
for ld in list_dict.itervalues()) # check length
# disassemble parallel expansion tuple and reassemble as individual arguments
if len(par_dict) > 0:
for fake, names in par_dict.iteritems():
assert fake in list_dict
par_args = zip(
*list_dict.pop(fake)
) # transpose, to get an expanded tuple for each argument
assert len(par_args) == len(names)
for name, args in zip(names, par_args):
list_dict[name] = args
# handle inner product expansion last
if len(inner_list) > 0:
kwtmp = kwargs.copy()
if len(outer_list) > 0:
kwtmp.update(list_dict)
inner_list = outer_list + inner_list
# N.B.: this replaces all outer expansion arguments with lists of appropriate length for inner expansion
inner_list, inner_dict = _prepareList(inner_list, kwtmp)
# inner product: essentially no expansion
lst0 = inner_dict[inner_list[0]]
lstlen = len(lst0)
for el in inner_list: # check length
if len(inner_dict[el]) == 1:
inner_dict[el] = inner_dict[
el] * lstlen # broadcast singleton list
elif len(inner_dict[el]) != lstlen:
raise TypeError(
'Lists have to be of same length to form inner product!'
)
list_dict = inner_dict
## generate list of argument dicts
arg_dicts = []
for n in xrange(lstlen):
# assemble arguments
lstargs = {key: lst[n] for key, lst in list_dict.iteritems()}
arg_dict = kwargs.copy()
arg_dict.update(lstargs)
arg_dicts.append(arg_dict)
# return list of arguments
return arg_dicts
def addDistFit(ensemble=None,
lfit=True,
lflatten=None,
lrescale=False,
reference=None,
target=None,
lbootstrap=False,
nbs=30,
sample_axis=None,
lglobalScale=False,
lcrossval=False,
ncv=0.2,
dist=None,
dist_args=None,
load_list=None,
lproduct='outer',
**kwargs):
''' add distribution fits to ensemble; optionally also rescale; kwargs are necessary for correct list expansion '''
# find appropriate sample axis
if lflatten:
if sample_axis is not None: raise ArgumentError, sample_axis
elif sample_axis is None: # auto-detect
for saxis in ('time', 'year'):
if all([all(ens.hasAxis(saxis)) for ens in ensemble]):
sample_axis = saxis
break
if sample_axis is None: raise AxisError, "No sample axis detected"
else:
if isinstance(sample_axis, basestring):
if not all([all(ens.hasAxis(sample_axis)) for ens in ensemble]):
raise AxisError, sample_axis
elif isinstance(sample_axis, (tuple, list)):
# check that axes are there
for ax in sample_axis:
if not all([all(ens.hasAxis(ax)) for ens in ensemble]):
raise AxisError, ax
# merge axes
ensemble = [ens.mergeAxes(axes=sample_axis, new_axis='sample', asVar=True, linplace=False, \
lcheckAxis=False) for ens in ensemble]
sample_axis = 'sample'
else:
raise AxisError, sample_axis
# perform fit or return dummy
if dist_args is None: dist_args = dict()
if lfit:
fitens = [
ens.fitDist(lflatten=lflatten,
axis=sample_axis,
lcrossval=lcrossval,
ncv=ncv,
lignoreParams=True,
lbootstrap=lbootstrap,
nbs=nbs,
dist=dist,
**dist_args) for ens in ensemble
]
else:
fitens = [None] * len(ensemble)
# rescale fitted distribution (according to certain rules)
if lrescale:
if not reference: raise ArgumentError(str(reference))
# expand target list
if isinstance(target, (list, tuple)):
expand_list = load_list[:]
if 'names' in expand_list:
expand_list[expand_list.index('names')] = 'target'
kwarg_list = expandArgumentList(target=target,
expand_list=expand_list,
lproduct=lproduct,
**kwargs)
targets = [kwarg['target'] for kwarg in kwarg_list]
else:
targets = [target] * len(fitens)
if isinstance(reference, (list, tuple)):
raise NotImplementedError # don't expand reference list
# use global reference, if necessary
if isinstance(reference, basestring) and not all(reference in fit
for fit in fitens):
i = 0
while i < len(fitens) and reference not in fitens[i]:
i += 1
if i >= len(fitens):
raise ArgumentError, "Reference {:s} not found in any dataset!".format(
reference)
reference = fitens[i][reference]
sclens = [
rescaleDistributions(fit,
reference=reference,
target=tgt,
lglobal=lglobalScale)
for fit, tgt in zip(fitens, targets)
]
else:
sclens = [None] * len(ensemble)
# return results
return fitens, sclens
def performExport(dataset, mode, dataargs, expargs, bcargs, loverwrite=False,
ldebug=False, lparallel=False, pidstr='', logger=None):
''' worker function to export ASCII rasters for a given dataset '''
# input checking
if not isinstance(dataset,basestring): raise TypeError
if not isinstance(dataargs,dict): raise TypeError # all dataset arguments are kwargs
# logging
if logger is None: # make new logger
logger = logging.getLogger() # new logger
logger.addHandler(logging.StreamHandler())
else:
if isinstance(logger,basestring):
logger = logging.getLogger(name=logger) # connect to existing one
elif not isinstance(logger,logging.Logger):
raise TypeError, 'Expected logger ID/handle in logger KW; got {}'.format(str(logger))
## extract meta data from arguments
dataargs, loadfct, srcage, datamsgstr = getMetaData(dataset, mode, dataargs, lone=False)
dataset_name = dataargs.dataset_name; periodstr = dataargs.periodstr; domain = dataargs.domain
# figure out bias correction parameters
if bcargs:
bcargs = bcargs.copy() # first copy, then modify...
bc_method = bcargs.pop('method',None)
if bc_method is None: raise ArgumentError("Need to specify bias-correction method to use bias correction!")
bc_obs = bcargs.pop('obs_dataset',None)
if bc_obs is None: raise ArgumentError("Need to specify observational dataset to use bias correction!")
bc_reference = bcargs.pop('reference',None)
if bc_reference is None: # infer from experiment name
if dataset_name[-5:] in ('-2050','-2100'): bc_reference = dataset_name[:-5] # cut of period indicator and hope for the best
else: bc_reference = dataset_name
bc_grid = bcargs.pop('grid',None)
if bc_grid is None: bc_grid = dataargs.grid
bc_domain = bcargs.pop('domain',None)
if bc_domain is None: bc_domain = domain
bc_varlist = bcargs.pop('varlist',None)
bc_varmap = bcargs.pop('varmap',None)
bc_tag = bcargs.pop('tag',None) # an optional name extension/tag
bc_pattern = bcargs.pop('file_pattern',None) # usually default in getPickleFile
lgzip = bcargs.pop('lgzip',None) # if pickle is gzipped (None: auto-detect based on file name extension)
# get name of pickle file (and folder)
picklefolder = dataargs.avgfolder.replace(dataset_name,bc_reference)
picklefile = getPickleFileName(method=bc_method, obs_name=bc_obs, gridstr=bc_grid, domain=bc_domain,
tag=bc_tag, pattern=bc_pattern)
picklepath = '{:s}/{:s}'.format(picklefolder,picklefile)
if lgzip:
picklepath += '.gz' # add extension
if not os.path.exists(picklepath): raise IOError(picklepath)
elif lgzip is None:
lgzip = False
if not os.path.exists(picklepath):
lgzip = True # assume gzipped file
picklepath += '.gz' # try with extension...
if not os.path.exists(picklepath): raise IOError(picklepath)
elif not os.path.exists(picklepath): raise IOError(picklepath)
pickleage = datetime.fromtimestamp(os.path.getmtime(picklepath))
# determine age of pickle file and compare against source age
else:
bc_method = False
pickleage = srcage
# parse export options
expargs = expargs.copy() # first copy, then modify...
lm3 = expargs.pop('lm3') # convert kg/m^2/s to m^3/m^2/s (water flux)
expformat = expargs.pop('format') # needed to get FileFormat object
exp_list= expargs.pop('exp_list') # this handled outside of export
compute_list = expargs.pop('compute_list', []) # variables to be (re-)computed - by default all
# initialize FileFormat class instance
fileFormat = getFileFormat(expformat, bc_method=bc_method, **expargs)
# get folder for target dataset and do some checks
expname = '{:s}_d{:02d}'.format(dataset_name,domain) if domain else dataset_name
expfolder = fileFormat.defineDataset(dataset=dataset, mode=mode, dataargs=dataargs, lwrite=True, ldebug=ldebug)
# prepare destination for new dataset
lskip = fileFormat.prepareDestination(srcage=max(srcage,pickleage), loverwrite=loverwrite)
# depending on last modification time of file or overwrite setting, start computation, or skip
if lskip:
# print message
skipmsg = "\n{:s} >>> Skipping: Format '{:s} for dataset '{:s}' already exists and is newer than source file.".format(pidstr,expformat,dataset_name)
skipmsg += "\n{:s} >>> ('{:s}')\n".format(pidstr,expfolder)
logger.info(skipmsg)
else:
## actually load datasets
source = loadfct() # load source data
# check period
if 'period' in source.atts and dataargs.periodstr != source.atts.period: # a NetCDF attribute
raise DateError, "Specifed period is inconsistent with netcdf records: '{:s}' != '{:s}'".format(periodstr,source.atts.period)
# load BiasCorrection object from pickle
if bc_method:
op = gzip.open if lgzip else open
with op(picklepath, 'r') as filehandle:
BC = pickle.load(filehandle)
# assemble logger entry
bcmsgstr = "(performing bias-correction using {:s} from {:s} towards {:s})".format(BC.long_name,bc_reference,bc_obs)
# print message
if mode == 'climatology': opmsgstr = 'Exporting Climatology ({:s}) to {:s} Format'.format(periodstr, expformat)
elif mode == 'time-series': opmsgstr = 'Exporting Time-series to {:s} Format'.format(expformat)
elif mode[-5:] == '-mean': opmsgstr = 'Exporting {:s}-Mean ({:s}) to {:s} Format'.format(mode[:-5], periodstr, expformat)
else: raise NotImplementedError, "Unrecognized Mode: '{:s}'".format(mode)
# print feedback to logger
logmsg = '\n{0:s} *** {1:^65s} *** \n{0:s} *** {2:^65s} *** \n'.format(pidstr,datamsgstr,opmsgstr)
if bc_method:
logmsg += "{0:s} *** {1:^65s} *** \n".format(pidstr,bcmsgstr)
logger.info(logmsg)
if not lparallel and ldebug: logger.info('\n'+str(source)+'\n')
# create GDAL-enabled target dataset
sink = Dataset(axes=(source.xlon,source.ylat), name=expname, title=source.title, atts=source.atts.copy())
addGDALtoDataset(dataset=sink, griddef=source.griddef)
assert sink.gdal, sink
# apply bias-correction
if bc_method:
source = BC.correct(source, asNC=False, varlist=bc_varlist, varmap=bc_varmap) # load bias-corrected variables into memory
# N.B.: for variables that are not bias-corrected, data are not loaded immediately but on demand; this way
# I/O and computing can be further disentangled and not all variables are always needed
# compute intermediate variables, if necessary
for varname in exp_list:
variables = None # variable list
var = None
# (re-)compute variable, if desired...
if varname in compute_list:
if varname == 'precip': var = newvars.computeTotalPrecip(source)
elif varname == 'waterflx': var = newvars.computeWaterFlux(source)
elif varname == 'liqwatflx': var = newvars.computeLiquidWaterFlux(source)
elif varname == 'netrad': var = newvars.computeNetRadiation(source, asVar=True)
elif varname == 'netrad_bb': var = newvars.computeNetRadiation(source, asVar=True, lrad=False, name='netrad_bb')
elif varname == 'netrad_bb0': var = newvars.computeNetRadiation(source, asVar=True, lrad=False, lA=False, name='netrad_bb0')
elif varname == 'vapdef': var = newvars.computeVaporDeficit(source)
elif varname in ('pet','pet_pm','petrad','petwnd') and 'pet' not in sink:
if 'petrad' in exp_list or 'petwnd' in exp_list:
variables = newvars.computePotEvapPM(source, lterms=True) # default; returns mutliple PET terms
else: var = newvars.computePotEvapPM(source, lterms=False) # returns only PET
elif varname == 'pet_th': var = None # skip for now
#var = computePotEvapTh(source) # simplified formula (less prerequisites)
# ... otherwise load from source file
if var is None and variables is None and varname in source:
var = source[varname].load() # load data (may not have to load all)
#else: raise VariableError, "Unsupported Variable '{:s}'.".format(varname)
# for now, skip variables that are None
if var or variables:
# handle lists as well
if var and variables: raise VariableError, (var,variables)
elif var: variables = (var,)
for var in variables:
addGDALtoVar(var=var, griddef=sink.griddef)
if not var.gdal and isinstance(fileFormat,ASCII_raster):
raise GDALError, "Exporting to ASCII_raster format requires GDAL-enabled variables."
# add to new dataset
sink += var
# convert units
if lm3:
for var in sink:
if var.units == 'kg/m^2/s':
var /= 1000. # divide to get m^3/m^2/s
var.units = 'm^3/m^2/s' # update units
# compute seasonal mean if we are in mean-mode
if mode[-5:] == '-mean':
sink = sink.seasonalMean(season=mode[:-5], lclim=True)
# N.B.: to remain consistent with other output modes,
# we need to prevent renaming of the time axis
sink = concatDatasets([sink,sink], axis='time', lensembleAxis=True)
sink.squeeze() # we need the year-axis until now to distinguish constant fields; now remove
# print dataset
if not lparallel and ldebug:
logger.info('\n'+str(sink)+'\n')
# export new dataset to selected format
fileFormat.exportDataset(sink)
# write results to file
writemsg = "\n{:s} >>> Export of Dataset '{:s}' to Format '{:s}' complete.".format(pidstr,expname, expformat)
writemsg += "\n{:s} >>> ('{:s}')\n".format(pidstr,expfolder)
logger.info(writemsg)
# clean up and return
source.unload(); #del source
return 0 # "exit code"
def __init__(self, inner_list=None, outer_list=None, **kwargs):
''' initialize an ensemble of HGS simulations based on HGS arguments and project descriptors;
all keyword arguments are automatically expanded based on inner/outer product rules, defined
using the inner_list/outer_list arguments; the expanded argument lists are used to initialize
the individual ensemble members; note that a string substitution is applied to all folder
variables (incl. 'rundir') prior to constructing the HGS instance, i.e. rundir.format(**kwargs) '''
self.lreport = kwargs.get('lreport', self.lreport)
self.loverwrite = kwargs.get('loverwrite', self.loverwrite)
self.lindicator = kwargs.get('lindicator', self.lindicator)
self.lrunfailed = kwargs.get('lrunfailed', self.lrunfailed)
self.lrestart = kwargs.get('lrestart', self.lrestart)
# expand argument list (plain, nothing special)
kwargs_list = expandArgumentList(inner_list=inner_list,
outer_list=outer_list,
**kwargs)
# loop over ensemble members
self.members = []
self.rundirs = []
self.hgsargs = [] # ensemble lists
for kwargs in kwargs_list:
# isolate folder variables and perform variable substitution
for folder_type in ('rundir', 'template_folder', 'input_folder',
'pet_folder', 'precip_inc', 'pet_inc',
'ic_files'):
if folder_type in kwargs:
folder = kwargs[folder_type]
if isinstance(folder, str):
# perform keyword substitution with all available arguments
if folder_type is 'ic_files':
# we need to preserve '{FILETYPE}' for later
kwargs[folder_type] = folder.format(
FILETYPE='{FILETYPE}', **kwargs)
else:
kwargs[folder_type] = folder.format(**kwargs)
elif folder is None:
pass
else:
raise TypeError(folder)
# check rundir
rundir = kwargs['rundir']
kwargs[
'restart'] = False # this keyword argument should be controlled by the Ensemble handler
if rundir in self.rundirs:
raise ArgumentError(
"Multiple occurence of run directory:\n '{}'".format(
rundir))
# figure out skipping
if os.path.exists(rundir):
if self.loverwrite:
if self.lreport:
print(
("Overwriting existing experiment folder '{:s}'.".
format(rundir)))
lskip = False
elif self.lindicator and os.path.exists(
'{}/SCHEDULED'.format(rundir)):
if self.lreport:
print(
("Skipping experiment folder '{:s}' (scheduled).".
format(rundir)))
lskip = True
elif self.lindicator and os.path.exists(
'{}/IN_PROGRESS'.format(rundir)):
if self.lrestart:
shutil.move(os.path.join(rundir, 'IN_PROGRESS'),
os.path.join(rundir, 'RESTARTED'))
if self.lreport:
print((
"Restarting experiment in folder '{:s}' (was in progress)."
.format(rundir)))
lskip = False
kwargs['restart'] = True
else:
if self.lreport:
print((
"Skipping experiment folder '{:s}' (in progress)."
.format(rundir)))
lskip = True
elif self.lindicator and os.path.exists(
'{}/COMPLETED'.format(rundir)):
if self.lreport:
print(
("Skipping experiment folder '{:s}' (completed).".
format(rundir)))
lskip = True
elif self.lindicator and os.path.exists(
'{}/FAILED'.format(rundir)):
# this should be the last option, so as to prevent overwriting data
if self.lrunfailed:
if self.lreport:
print(
("Overwriting failed experiment folder '{:s}'."
.format(rundir)))
lskip = False # rundir will be deleted
else:
if self.lreport:
print(
("Skipping experiment folder '{:s}' (failed).".
format(rundir)))
lskip = True
else: # no/unknown indicator file
if self.lreport:
print(
("Overwriting existing experiment folder '{:s}'.".
format(rundir)))
lskip = False # rundir will be deleted
else:
if self.lreport:
print(("Creating new experiment folder '{:s}'.".format(
rundir)))
lskip = False
if not lskip:
self.rundirs.append(rundir)
# isolate HGS constructor arguments
hgsargs = inspect.getargspec(
HGS.__init__
).args # returns args, varargs, kwargs, defaults
hgsargs = {
arg: kwargs[arg]
for arg in hgsargs if arg in kwargs
}
self.hgsargs.append(hgsargs)
# initialize HGS instance
hgs = HGS(**hgsargs)
self.members.append(hgs)
# final check
if len(self.members) == 0:
raise EnsembleError("No experiments to run (empty list).")
