def generate_residual_IDR2_2(uvh5_file,
omni_vis,
omni_calfits,
abs_calfits,
outfile,
clobber=False):
# reading uvh5 data file
hd = HERAData(uvh5_file)
data, flags, nsamples = hd.read(polarizations=['ee', 'nn'])
# reading omnical model visibilities
hd_oc = HERAData(omni_vis)
omnivis, omnivis_flags, _ = hd_oc.read()
uvo = pyuvdata.UVData()
uvo.read_uvh5(omni_vis)
# reading calfits file
hc = HERACal(omni_calfits)
oc_gains, oc_flags, oc_quals, oc_total_quals = hc.read()
hc = HERACal(abs_calfits)
ac_gains, ac_flags, ac_quals, ac_total_quals = hc.read()
# calibrating the data
abscal_data, abscal_flags = copy.deepcopy(data), copy.deepcopy(flags)
calibrate_in_place(abscal_data,
ac_gains,
data_flags=abscal_flags,
cal_flags=ac_flags)
res_data, res_flags = copy.deepcopy(hd.data_array), copy.deepcopy(
hd.flag_array)
resdata, resflags = copy.deepcopy(abscal_data), copy.deepcopy(abscal_flags)
for i, p in enumerate(['ee', 'nn']):
# reading omnical model visibilities
hd_oc = HERAData(omni_vis)
omnivis, omnivis_flags, _ = hd_oc.read(polarizations=[p])
mod_bls = list(omnivis.keys())
red_bls = get_reds(hd.antpos, pols=p)
red = gr.RBL(red_bls)
for mbl in mod_bls:
bl_grp = red[tuple(mbl[0:2]) + ('J{}'.format(p), )]
for blp in bl_grp:
bl = (blp[0], blp[1], p)
inds = hd.antpair2ind(bl)
omnivis_scaled = omnivis[mbl] * oc_gains[(blp[0], 'J{}'.format(
p))] * np.conj(oc_gains[(blp[1], 'J{}'.format(p))])
omnivis_scaled /= (
ac_gains[(blp[0], 'J{}'.format(p))] *
np.conj(ac_gains[(blp[1], 'J{}'.format(p))]))
resdata[bl] = abscal_data[bl] - omnivis_scaled
resflags[bl] = abscal_flags[bl]
res_data[inds, 0, :, i] = resdata[bl]
res_flags[inds, 0, :, i] = resflags[bl]
# writing to file
hd.data_array = res_data
hd.flag_array = res_flags
hd.write_uvh5(outfile, clobber=clobber)
def test_init_HERAData():
uv = UVData()
uv.read_miriad(test_d_file)
uvf1 = UVFlag(uv)
hd = HERAData(test_d_file, filetype='miriad')
hd.read()
uvf2 = UVFlag(hd)
nt.assert_equal(uvf1, uvf2)
def reds_from_file(filename, vis_format='miriad'):
"""Get the redundant baseline pairs from a file.
This is a wrapper around hera_cal.redcal.get_pos_reds that doesn't read
the data file if it's possible to only read metadata.
Parameters
----------
filename : str
The file to get reds from.
vis_format : {'miriad', 'uvh5', 'uvfits', 'fhd', 'ms'}, optional
Format of the data file. Default is 'miriad'.
Returns
-------
reds : list of lists of tuples
Each tuple represents antenna pairs. These are compiled in a list within
a redundant group, and the outer list is all the redundant groups.
See hera_cal.redcal.get_pos_reds.
"""
from hera_cal.io import HERAData
from hera_cal.redcal import get_pos_reds
hd = HERAData(filename, filetype=vis_format)
if hd.antpos is None:
reds = get_pos_reds(hd.read()[0].antpos)
else:
reds = get_pos_reds(hd.antpos)
del hd
return reds
def test_reds_from_file_read_file():
from hera_cal.io import HERAData
from hera_cal.redcal import get_pos_reds
# Miriad file will need to be read in
testfile = os.path.join(DATA_PATH, 'zen.2457698.40355.xx.HH.uvcAA')
reds = ant_metrics.reds_from_file(testfile, vis_format='miriad')
assert len(reds) > 1
hd = HERAData(testfile, filetype='miriad')
reds_check = get_pos_reds(hd.read()[0].antpos)
assert reds == reds_check
class AntennaMetrics():
"""Container for holding data and meta-data for ant metrics calculations.
This class creates an object for holding relevant visibility data and metadata,
and provides interfaces to four antenna metrics: two identify dead antennas,
and two identify cross-polarized antennas. These metrics can be used iteratively
to identify bad antennas. The object handles all stroage of metrics, and supports
writing metrics to an HDF5 filetype. The analysis functions are designed to work
on raw data from a single observation with all four polarizations.
"""
def __init__(self, dataFileList, reds, fileformat='miriad'):
"""Initilize an AntennaMetrics object.
Parameters
----------
dataFileList : list of str
List of data filenames of the four different visibility polarizations
for the same observation.
reds : list of tuples of ints
List of lists of tuples of antenna numbers that make up redundant baseline groups.
format : str, optional
File type of data. Must be one of: 'miriad', 'uvh5', 'uvfits', 'fhd',
'ms' (see pyuvdata docs). Default is 'miriad'.
Attributes
----------
hd : HERAData
HERAData object generated from dataFileList.
data : array
Data contained in HERAData object.
flags : array
Flags contained in HERAData object.
nsamples : array
Nsamples contained in HERAData object.
ants : list of ints
List of antennas in HERAData object.
pols : list of str
List of polarizations in HERAData object.
bls : list of ints
List of baselines in HERAData object.
dataFileList : list of str
List of data filenames of the four different visibility polarizations
for the same observation.
reds : list of tuples of ints
List of lists of tuples of antenna numbers that make up redundant baseline groups.
version_str : str
The version of the hera_qm module used to generate these metrics.
history : str
History to append to the metrics files when writing out files.
"""
from hera_cal.io import HERAData
self.hd = HERAData(dataFileList, filetype=fileformat)
self.data, self.flags, self.nsamples = self.hd.read()
self.ants = self.hd.get_ants()
self.pols = [pol.lower() for pol in self.hd.get_pols()]
self.antpols = [antpol.lower() for antpol in self.hd.get_feedpols()]
self.bls = self.hd.get_antpairs()
self.dataFileList = dataFileList
self.reds = reds
self.version_str = hera_qm_version_str
self.history = ''
if len(self.antpols) != 2 or len(self.pols) != 4:
raise ValueError('Missing polarization information. pols ='
+ str(self.pols) + ' and antpols = '
+ str(self.antpols))
def mean_Vij_metrics(self, pols=None, xants=[], rawMetric=False):
"""Calculate how an antennas's average |Vij| deviates from others.
Local wrapper for mean_Vij_metrics in hera_qm.ant_metrics module
Parameters
----------
pols : list of str, optional
List of visibility polarizations (e.g. ['xx','xy','yx','yy']).
Default is self.pols.
xants : list of tuples, optional
List of antenna-polarization tuples that should be ignored. The
expected format is (ant, antpol). Default is empty list.
rawMetric : bool, optional
If True, return the raw mean Vij metric instead of the modified z-score.
Default is False.
Returns
-------
meanMetrics : dict
Dictionary indexed by (ant, antpol) keys. Contains the modified z-score
of the mean of the absolute value of all visibilities associated with
an antenna. Very small or very large numbers are probably bad antennas.
"""
if pols is None:
pols = self.pols
return mean_Vij_metrics(self.data, pols, self.antpols,
self.ants, self.bls, xants=xants,
rawMetric=rawMetric)
def red_corr_metrics(self, pols=None, xants=[], rawMetric=False,
crossPol=False):
"""Calculate modified Z-Score over all redundant groups for each antenna.
This method is a local wrapper for red_corr_metrics. It calculates the extent
to which baselines involving an antenna do not correlate with others they are
nominmally redundant with.
Parameters
----------
data : array or HERAData
Data for all polarizations, stored in a format that supports indexing
as data[i,j,pol].
pols : list of str, optional
List of visibility polarizations (e.g. ['xx','xy','yx','yy']).
Default is self.pols.
xants : list of tuples, optional
List of antenna-polarization tuples that should be ignored. The
expected format is (ant, antpol). Default is empty list.
rawMetric : bool, optional
If True, return the raw power correlations instead of the modified z-score.
Default is False.
crossPol : bool, optional
If True, return results only when the two visibility polarizations differ
by a single flip. Default is False.
Returns
-------
powerRedMetric : dict
Dictionary indexed by (ant,antpol) keys. Contains the modified z-scores
of the mean power correlations inside redundant baseline groups associated
with each antenna. Very small numbers are probably bad antennas.
"""
if pols is None:
pols = self.pols
return red_corr_metrics(self.data, pols, self.antpols,
self.ants, self.reds, xants=xants,
rawMetric=rawMetric, crossPol=crossPol)
def mean_Vij_cross_pol_metrics(self, xants=[], rawMetric=False):
"""Calculate the ratio of cross-pol visibilities to same-pol visibilities.
This method is a local wrapper for mean_Vij_cross_pol_metrics. It finds
which antennas are outliers based on the ratio of mean cross-pol visibilities
to mean same-pol visibilities:
(Vxy+Vyx)/(Vxx+Vyy).
Parameters
----------
xants : list of tuples, optional
List of antenna-polarization tuples that should be ignored. The
expected format is (ant, antpol). Default is empty list.
rawMetric : bool, optional
If True, return the raw power correlations instead of the modified z-score.
Default is False.
Returns
-------
mean_Vij_cross_pol_metrics : dict
Dictionary indexed by (ant,antpol) keys. Contains the modified z-scores of the
ratio of mean visibilities, (Vxy+Vyx)/(Vxx+Vyy). Results are duplicated in
both antpols. Very large values are likely cross-polarized.
"""
return mean_Vij_cross_pol_metrics(self.data, self.pols,
self.antpols, self.ants,
self.bls, xants=xants,
rawMetric=rawMetric)
def red_corr_cross_pol_metrics(self, xants=[], rawMetric=False):
"""Calculate modified Z-Score over redundant groups; assume cross-polarized.
This method is a local wrapper for red_corr_cross_pol_metrics. It finds
which antennas are part of visibilities that are significantly better
correlated with polarization-flipped visibilities in a redundant group.
It returns the modified z-score.
Parameters
----------
xants : list of tuples, optional
List of antenna-polarization tuples that should be ignored. The
expected format is (ant, antpol). Default is empty list.
rawMetric : bool, optional
If True, return the raw power correlations instead of the modified z-score.
Default is False.
Returns
-------
redCorrCrossPolMetrics : dict
Dictionary indexed by (ant,antpol) keys. Contains the modified z-scores
of the mean correlation ratio between redundant visibilities and singly-
polarization flipped ones. Very large values are probably cross-polarized.
"""
return red_corr_cross_pol_metrics(self.data, self.pols,
self.antpols, self.ants,
self.reds, xants=xants,
rawMetric=rawMetric)
def reset_summary_stats(self):
"""Reset all the internal summary statistics back to empty."""
self.xants, self.crossedAntsRemoved, self.deadAntsRemoved = [], [], []
self.iter = 0
self.removalIter = {}
self.allMetrics, self.allModzScores = OrderedDict(), OrderedDict()
self.finalMetrics, self.finalModzScores = {}, {}
def find_totally_dead_ants(self):
"""Flag antennas whose median autoPower is 0.0.
These antennas are marked as dead. They do not appear in recorded antenna
metrics or zscores. Their removal iteration is -1 (i.e. before iterative
flagging).
"""
autoPowers = compute_median_auto_power_dict(self.data,
self.pols,
self.reds)
power_list_by_ant = {(ant, antpol): []
for ant in self.ants
for antpol in self.antpols
if (ant, antpol) not in self.xants}
for ((ant0, ant1, pol), power) in autoPowers.items():
if ((ant0, pol[0]) not in self.xants
and (ant1, pol[1]) not in self.xants):
power_list_by_ant[(ant0, pol[0])].append(power)
power_list_by_ant[(ant1, pol[1])].append(power)
for (key, val) in power_list_by_ant.items():
if np.median(val) == 0:
self.xants.append(key)
self.deadAntsRemoved.append(key)
self.removalIter[key] = -1
def _run_all_metrics(self, run_mean_vij=True, run_red_corr=True,
run_cross_pols=True, run_cross_pols_only=False):
"""Local call for all metrics as part of iterative flagging method.
Parameters
----------
run_mean_vij : bool, optional
Define if mean_Vij_metrics or mean_Vij_cross_pol_metrics are executed.
Default is True.
run_red_corr : bool, optional
Define if red_corr_metrics or red_corr_cross_pol_metrics are executed.
Default is True.
run_cross_pols : bool, optional
Define if mean_Vij_cross_pol_metrics and red_corr_cross_pol_metrics
are executed. Default is True. Individual rules are inherited from
run_mean_vij and run_red_corr.
run_cross_pols_only : bool, optional
Define if cross pol metrics are the *only* metrics to be run.
Default is False.
"""
# Compute all raw metrics
metNames = []
metVals = []
if run_mean_vij and not run_cross_pols_only:
metNames.append('meanVij')
meanVij = self.mean_Vij_metrics(pols=self.pols,
xants=self.xants,
rawMetric=True)
metVals.append(meanVij)
if run_red_corr and not run_cross_pols_only:
metNames.append('redCorr')
pols = [pol for pol in self.pols if pol[0] == pol[1]]
redCorr = self.red_corr_metrics(pols=pols,
xants=self.xants,
rawMetric=True)
metVals.append(redCorr)
if run_cross_pols:
if run_mean_vij:
metNames.append('meanVijXPol')
meanVijXPol = self.mean_Vij_cross_pol_metrics(xants=self.xants,
rawMetric=True)
metVals.append(meanVijXPol)
if run_red_corr:
metNames.append('redCorrXPol')
redCorrXPol = self.red_corr_cross_pol_metrics(xants=self.xants,
rawMetric=True)
metVals.append(redCorrXPol)
# Save all metrics and zscores
metrics, modzScores = {}, {}
for metric, metName in zip(metVals, metNames):
metrics[metName] = metric
modz = per_antenna_modified_z_scores(metric)
modzScores[metName] = modz
for key in metric:
if metName in self.finalMetrics:
self.finalMetrics[metName][key] = metric[key]
self.finalModzScores[metName][key] = modz[key]
else:
self.finalMetrics[metName] = {key: metric[key]}
self.finalModzScores[metName] = {key: modz[key]}
self.allMetrics.update({self.iter: metrics})
self.allModzScores.update({self.iter: modzScores})
def iterative_antenna_metrics_and_flagging(self, crossCut=5, deadCut=5,
alwaysDeadCut=10,
verbose=False,
run_mean_vij=True,
run_red_corr=True,
run_cross_pols=True,
run_cross_pols_only=False):
"""Run all four antenna metrics and stores results in self.
Runs all four metrics: two for dead antennas, two for cross-polarized antennas.
Saves the results internally to this this antenna metrics object.
Parameters
----------
crossCut : float, optional
Modified z-score cut for most cross-polarized antennas. Default is 5 "sigmas".
deadCut : float, optional
Modified z-score cut for most likely dead antennas. Default is 5 "sigmas".
alwaysDeadCut : float, optional
Modified z-score cut for definitely dead antennas. Default is 10 "sigmas".
These are all thrown away at once without waiting to iteratively throw away
only the worst offender.
run_mean_vij : bool, optional
Define if mean_Vij_metrics or mean_Vij_cross_pol_metrics are executed.
Default is True.
run_red_corr : bool, optional
Define if red_corr_metrics or red_corr_cross_pol_metrics are executed.
Default is True.
run_cross_pols : bool, optional
Define if mean_Vij_cross_pol_metrics and red_corr_cross_pol_metrics
are executed. Default is True. Individual rules are inherited from
run_mean_vij and run_red_corr.
run_cross_pols_only : bool, optional
Define if cross pol metrics are the *only* metrics to be run. Default
is False.
"""
self.reset_summary_stats()
self.find_totally_dead_ants()
self.crossCut, self.deadCut = crossCut, deadCut
self.alwaysDeadCut = alwaysDeadCut
# Loop over
for iter in range(len(self.antpols) * len(self.ants)):
self.iter = iter
self._run_all_metrics(run_mean_vij=run_mean_vij,
run_red_corr=run_red_corr,
run_cross_pols=run_cross_pols,
run_cross_pols_only=run_cross_pols_only)
# Mostly likely dead antenna
last_iter = list(self.allModzScores)[-1]
worstDeadCutRatio = -1
worstCrossCutRatio = -1
if run_mean_vij and run_red_corr and not run_cross_pols_only:
deadMetrics = average_abs_metrics(self.allModzScores[last_iter]['meanVij'],
self.allModzScores[last_iter]['redCorr'])
else:
if run_mean_vij and not run_cross_pols_only:
deadMetrics = self.allModzScores[last_iter]['meanVij'].copy()
elif run_red_corr and not run_cross_pols_only:
deadMetrics = self.allModzScores[last_iter]['redCorr'].copy()
try:
worstDeadAnt = max(deadMetrics, key=deadMetrics.get)
worstDeadCutRatio = np.abs(deadMetrics[worstDeadAnt]) / deadCut
except NameError:
# Dead metrics weren't run, but that's fine.
pass
if run_cross_pols:
# Most likely cross-polarized antenna
if run_mean_vij and run_red_corr:
crossMetrics = average_abs_metrics(self.allModzScores[last_iter]['meanVijXPol'],
self.allModzScores[last_iter]['redCorrXPol'])
elif run_mean_vij:
crossMetrics = self.allModzScores[last_iter]['meanVijXPol'].copy()
elif run_red_corr:
crossMetrics = self.allModzScores[last_iter]['redCorrXPol'].copy()
worstCrossAnt = max(crossMetrics, key=crossMetrics.get)
worstCrossCutRatio = (np.abs(crossMetrics[worstCrossAnt])
/ crossCut)
# Find the single worst antenna, remove it, log it, and run again
if (worstCrossCutRatio >= worstDeadCutRatio
and worstCrossCutRatio >= 1.0):
for antpol in self.antpols:
self.xants.append((worstCrossAnt[0], antpol))
self.crossedAntsRemoved.append((worstCrossAnt[0], antpol))
self.removalIter[(worstCrossAnt[0], antpol)] = iter
if verbose:
print('On iteration', iter, 'we flag\t', end='')
print((worstCrossAnt[0], antpol))
elif (worstDeadCutRatio > worstCrossCutRatio
and worstDeadCutRatio > 1.0):
dead_ants = set([worstDeadAnt])
for (ant, metric) in deadMetrics.items():
if metric > alwaysDeadCut:
dead_ants.add(ant)
for dead_ant in dead_ants:
self.xants.append(dead_ant)
self.deadAntsRemoved.append(dead_ant)
self.removalIter[dead_ant] = iter
if verbose:
print('On iteration', iter, 'we flag', dead_ant)
else:
break
def save_antenna_metrics(self, filename, overwrite=False):
"""Output all meta-metrics and cut decisions to HDF5 file.
Saves all cut decisions and meta-metrics in an HDF5 that can be loaded
back into a dictionary using hera_qm.ant_metrics.load_antenna_metrics()
Parameters
----------
filename : str
The file into which metrics will be written.
overwrite: bool, optional
Whether to overwrite an existing file. Default is False.
"""
if not hasattr(self, 'xants'):
raise KeyError(('Must run AntennaMetrics.'
'iterative_antenna_metrics_and_flagging() first.'))
out_dict = {'xants': self.xants}
out_dict['crossed_ants'] = self.crossedAntsRemoved
out_dict['dead_ants'] = self.deadAntsRemoved
out_dict['final_metrics'] = self.finalMetrics
out_dict['all_metrics'] = self.allMetrics
out_dict['final_mod_z_scores'] = self.finalModzScores
out_dict['all_mod_z_scores'] = self.allModzScores
out_dict['removal_iteration'] = self.removalIter
out_dict['cross_pol_z_cut'] = self.crossCut
out_dict['dead_ant_z_cut'] = self.deadCut
out_dict['always_dead_ant_z_cut'] = self.alwaysDeadCut
out_dict['datafile_list'] = self.dataFileList
out_dict['reds'] = self.reds
metrics_io.write_metric_file(filename, out_dict, overwrite=overwrite)
model_file = model_directory
flag_files = [f"/lustre/aoc/projects/hera/aewallwi/H1C_flags/{jd}.flags.h5" for jd in [2458098,2458099,2458101,2458102,2458103,2458104,2458105,2458106,
2458107,2458108,2458109,2458110,2458111,2458112,2458113,2458114,
2458115,2458116]]
bad_ants = [np.loadtxt(f"/users/kshahin/kshahin/HERA_Calibration/hera_pipelines/pipelines/h1c/idr2/v2/bad_ants/{ba}.txt") for ba in [2458098,2458099,2458101,2458102,2458103,2458104,2458105,2458106,2458107,2458108,
2458109,2458110,2458111,2458112,2458113,2458114,2458115,2458116]]
flag_file = flag_files[day]
bad_ant = bad_ants[day]
if not os.path.exists(f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}"):
os.mkdir(f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}")
flags = UVFlag(flag_file)
flags.select(frequencies = flags.freq_array[(flags.freq_array>=115*1e+6) & (flags.freq_array<175*1e+6)])
flags.select(times=flags.time_array[2600:2660])
hd_data = HERAData(data_file)
freqs = hd_data.freqs[(hd_data.freqs>=115*1e+6) & (hd_data.freqs<175*1e+6)]
data, flag, nsample = hd_data.read(polarizations=["ee"], frequencies=freqs)
for bl in data:
if (bl[0] == bad_ant).any() or (bl[1] == bad_ant).any():
flag[bl] = np.ones_like(flag[bl])
flag[bl] = flags.flag_array.squeeze()
hd_data.update(flags=flag)
hd_data.write_uvh5(f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}/data_{day}_{chunk}.uvh5", clobber=True)
del data, flag, nsample, hd_data
redcal.redcal_run(input_data=f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}/data_{day}_{chunk}.uvh5", clobber=True, solar_horizon=90, verbose=True)
abscal.post_redcal_abscal_run(data_file=f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}/data_{day}_{chunk}.uvh5", redcal_file=f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}/data_{day}_{chunk}.omni.calfits", model_files=[data_file], clobber=True, data_solar_horizon=90, model_solar_horizon=90)
cs=smooth_cal.CalibrationSmoother(calfits_list=sorted(glob.glob(f"/users/kshahin/kshahin/HERA_Calibration/DayfxP_{day}/data_{day}_*.abs.calfits")))
cs.time_freq_2D_filter(time_scale=21600)
cs.write_smoothed_cal(clobber=True, output_replace=(".abs.",".smooth_abs."))
def ant_metrics_run(files, pols=['xx', 'yy', 'xy', 'xy'], crossCut=5.0,
deadCut=5.0, alwaysDeadCut=10.0, metrics_path='',
extension='.ant_metrics.hdf5', vis_format='miriad',
verbose=True, history='',
run_mean_vij=True, run_red_corr=True,
run_cross_pols=True):
"""
Run a series of ant_metrics tests on a given set of input files.
Args:
files: List of files to run ant metrics on.
Can be any of the 4 polarizations
pols: List of polarizations to perform metrics over.
Allowed polarizations: 'xx', 'yy', 'xy', 'yx'
Default: ['xx', 'yy', 'xy', 'yx']
crossCut: Modified Z-Score limit to cut cross-polarized antennas.
Default: 5.0
deadCut: Modifized Z-Score limit to cut dead antennas.
Default: 5.0
alwaysDeadCut: Modified Z-Score limit for antennas that are definitely dead.
Antennas with above this limit are thrown away before iterative flagging.
Default: 10.0
metrics_path: String path to directory to story output metric.
Default: same directy as input data files.
extension: File extension to add to output files.
Default: ant_metrics.hdf5
vis_format: File format of input visibility data.
Supports: 'miriad','uvfits', 'fhd', 'ms' (see pyuvdata docs)
Default: 'miriad'
verbose: If True print out statements during iterative flagging
history: The history the add to metrics.
Default
run_mean_vij: Boolean flag which determines if mean_Vij_metrics is executed.
Default is True
run_red_corr: Boolean flag which determines if red_corr_metrics is executed.
Default is True
run_cross_pols: Boolean flag which determines if mean_Vij_cross_pol_metrics and red_corr_cross_pol_metrics are executed.
Default is True
Return:
None
The funciton will take in a list of files and options.
It will run the series of ant metrics tests,
and produce an HDF5 file containing the relevant information.
The file list need only contain one polarization type for a given JD,
the function will look for the other polarizations in the same folder.
If not all four polarizations are found, a warning is
generated, since the code assumes all four polarizations are present.
"""
from hera_cal.omni import aa_to_info
from hera_cal.utils import get_aa_from_uv
from hera_cal.io import HERAData
# check the user asked to run anything
if not any([run_mean_vij, run_red_corr, run_cross_pols]):
raise AssertionError(("No Ant Metrics have been selected to run."
"Please set the correct keywords to run "
"the desired metrics."))
# check that we were given some files to process
if len(files) == 0:
raise AssertionError('Please provide a list of visibility files')
# generate a list of all files to be read in
fullpol_file_list = utils.generate_fullpol_file_list(files, pols)
if len(fullpol_file_list) == 0:
raise AssertionError('Could not find all 4 polarizations '
'for any files provided')
# generate aa object from file
# N.B.: assumes redunancy information is the same for all files passed in
first_file = fullpol_file_list[0][0]
hd = HERAData(first_file, filetype='miriad')
data, flags, nsamples = hd.read()
aa = get_aa_from_uv(hd)
del hd
info = aa_to_info(aa, pols=[pols[-1][0]])
reds = info.get_reds()
# do the work
for jd_list in fullpol_file_list:
am = AntennaMetrics(jd_list, reds, fileformat=vis_format)
am.iterative_antenna_metrics_and_flagging(crossCut=crossCut,
deadCut=deadCut,
alwaysDeadCut=alwaysDeadCut,
verbose=verbose,
run_mean_vij=run_mean_vij,
run_red_corr=run_red_corr,
run_cross_pols=run_cross_pols)
# add history
am.history = am.history + history
base_filename = jd_list[0]
abspath = os.path.abspath(base_filename)
dirname = os.path.dirname(abspath)
basename = os.path.basename(base_filename)
nopol_filename = re.sub('\.{}\.'.format(pols[0]), '.', basename)
if metrics_path == '':
# default path is same directory as file
metrics_path = dirname
else:
metrics_path = metrics_path
metrics_basename = nopol_filename + extension
metrics_filename = os.path.join(metrics_path, metrics_basename)
am.save_antenna_metrics(metrics_filename)
return
class AntennaMetrics():
"""Container for holding data and meta-data for ant metrics calculations.
Object for holding relevant visibility data and metadata
interfaces to four antenna metrics:
Two each for identifying dead or cross-polarized antennas.
Can iteratively identifying bad antennas
Handles all metrics stroage, and supports writing metrics to HDF5.
Works on raw data from a single observation with all four polarizations.
"""
def __init__(self, dataFileList, reds, fileformat='miriad'):
"""Initilize an AntennaMetrics object.
Arguments:
dataFileList: List of data filenames of the four different visibility polarizations for the same observation
reds: List of lists of tuples of antenna numbers that make up redundant baseline groups.
format: File type of data
Supports: 'miriad','uvfits', 'fhd', 'ms ' (see pyuvdata docs)
Default: 'miriad'.
"""
from hera_cal.io import HERAData
if fileformat == 'miriad':
self.hd = HERAData(dataFileList, filetype='miriad')
elif fileformat == 'uvfits':
self.hd = HERAData(dataFileList, filetype='uvfits')
elif fileformat == 'fhd':
raise NotImplemented(str(fileformat) + 'not supported')
else:
raise ValueError('Unrecognized file format ' + str(fileformat))
self.data, self.flags, self.nsamples = self.hd.read()
self.ants = self.hd.get_ants()
self.pols = [pol.lower() for pol in self.hd.get_pols()]
self.antpols = [antpol.lower() for antpol in self.hd.get_feedpols()]
self.bls = self.hd.get_antpairs()
self.dataFileList = dataFileList
self.reds = reds
self.version_str = hera_qm_version_str
self.history = ''
if len(self.antpols) is not 2 or len(self.pols) is not 4:
raise ValueError('Missing polarization information. pols ='
+ str(self.pols) + ' and antpols = '
+ str(self.antpols))
def mean_Vij_metrics(self, pols=None, xants=[], rawMetric=False):
"""Calculate how an antennas's average |Vij| deviates from others.
Local wrapper for mean_Vij_metrics in hera_qm.ant_metrics module
Arguments:
pols : List of visibility polarizations (e.g. ['xx','xy','yx','yy'])
Default: self.pols
xants: List of antennas ithat should be ignored.
format: (ant,antpol)
rawMetric:return the raw mean Vij metric instead of the modified z-score
Returns:
meanMetrics: Dictionary indexed by (ant,antpol) of the modified z-score
of the mean of the absolute value of all visibilities associated with an antenna.
Very small or very large numbers are probably bad antennas.
"""
if pols is None:
pols = self.pols
return mean_Vij_metrics(self.data, pols, self.antpols,
self.ants, self.bls, xants=xants,
rawMetric=rawMetric)
def red_corr_metrics(self, pols=None, xants=[], rawMetric=False,
crossPol=False):
"""Calculate modified Z-Score over all redundant groups for each antenna.
Local wrapper for red_corr_metrics in hera_qm.ant_metrics module.
Calculates the extent to which baselines involving an antenna
do not correlate with others they are nominmally redundant with.
Arguments:
data: data for all polarizations
format must support data[i,j,pol]
pols: List of visibility polarizations (e.g. ['xx','xy','yx','yy']).
Default: self.pols
xants: List of antennas that should be ignored.
format: (ant, antpol)
rawMetric: return the raw power correlations instead of the modified z-score
crossPol: return results only when the two visibility polarizations differ by a single flip
Returns:
powerRedMetric: Dictionary indexed by (ant,antpol)
of the modified z-scores of the mean power correlations
inside redundant baseline groups
associated with each antenna.
Very small numbers are probably bad antennas.
"""
if pols is None:
pols = self.pols
return red_corr_metrics(self.data, pols, self.antpols,
self.ants, self.reds, xants=xants,
rawMetric=rawMetric, crossPol=crossPol)
def mean_Vij_cross_pol_metrics(self, xants=[], rawMetric=False):
"""Calculate the ratio of cross-pol visibilities to same-pol visibilities.
Local wrapper for mean_Vij_cross_pol_metrics.
Find which antennas are outliers based on the
ratio of mean cross-pol visibilities to mean same-pol visibilities:
(Vxy+Vyx)/(Vxx+Vyy).
Arguments:
xants: List of antennas that should be ignored.
format (ant,antpol) format
e.g., if (81,'y') is excluded, (81,'x') cannot be identified
as cross-polarized and will be excluded.
rawMetric: return the raw power ratio instead of the modified z-score
Returns:
mean_Vij_cross_pol_metrics: Dictionary indexed by (ant,antpol)
The modified z-scores of the
ratio of mean visibilities,
(Vxy+Vyx)/(Vxx+Vyy).
Results duplicated in both antpols.
Very large values are likely cross-polarized.
"""
return mean_Vij_cross_pol_metrics(self.data, self.pols,
self.antpols, self.ants,
self.bls, xants=xants,
rawMetric=rawMetric)
def red_corr_cross_pol_metrics(self, xants=[], rawMetric=False):
"""Calculate modified Z-Score over redundant groups; assume cross-polarized.
Local wrapper for red_corr_cross_pol_metrics.
Find which antennas are part of visibilities that are significantly better
correlated with polarization-flipped visibilities in a redundant groupself.
Returns the modified z-score.
Arguments:
xants: List of antennas that should be ignored.
format (ant,antpol) format
e.g., if (81,'y') is excluded, (81,'x') cannot be identified
as cross-polarized and will be excluded.
rawMetric: return the raw power ratio instead of the modified z-score
type: Boolean
Default: False
Returns:
redCorrCrossPolMetrics: Dictionary indexed by (ant,antpol)
The modified z-scores of the mean correlation
ratio between redundant visibilities
and singlely-polarization flipped ones.
Very large values are probably cross-polarized.
"""
return red_corr_cross_pol_metrics(self.data, self.pols,
self.antpols, self.ants,
self.reds, xants=xants,
rawMetric=rawMetric)
def reset_summary_stats(self):
"""Reset all the internal summary statistics back to empty."""
self.xants, self.crossedAntsRemoved, self.deadAntsRemoved = [], [], []
self.iter = 0
self.removalIter = {}
self.allMetrics, self.allModzScores = OrderedDict(), OrderedDict()
self.finalMetrics, self.finalModzScores = {}, {}
def find_totally_dead_ants(self):
"""Flag antennas whose median autoPower is 0.0.
These antennas are marked as dead
They do not appear in recorded antenna metrics or zscores.
Their removal iteration is -1 (i.e. before iterative flagging).
"""
autoPowers = compute_median_auto_power_dict(self.data,
self.pols,
self.reds)
power_list_by_ant = {(ant, antpol): []
for ant in self.ants
for antpol in self.antpols
if (ant, antpol) not in self.xants}
for ((ant0, ant1, pol), power) in autoPowers.items():
if ((ant0, pol[0]) not in self.xants
and (ant1, pol[1]) not in self.xants):
power_list_by_ant[(ant0, pol[0])].append(power)
power_list_by_ant[(ant1, pol[1])].append(power)
for (key, val) in power_list_by_ant.items():
if np.median(val) == 0:
self.xants.append(key)
self.deadAntsRemoved.append(key)
self.removalIter[key] = -1
def _run_all_metrics(self, run_mean_vij=True, run_red_corr=True,
run_cross_pols=True):
"""Local call for all metrics as part of iterative flagging method.
Arguments:
run_mean_vij: Boolean flag which determines if mean_Vij_metrics is executed.
Default is True
run_red_corr: Boolean flag which determines if red_corr_metrics is executed.
Default is True
run_cross_pols: Boolean flag which determines if mean_Vij_cross_pol_metrics and red_corr_cross_pol_metrics are executed.
Default is True
"""
# Compute all raw metrics
metNames = []
metVals = []
if run_mean_vij:
metNames.append('meanVij')
meanVij = self.mean_Vij_metrics(pols=self.pols,
xants=self.xants,
rawMetric=True)
metVals.append(meanVij)
if run_red_corr:
metNames.append('redCorr')
redCorr = self.red_corr_metrics(pols=['xx', 'yy'],
xants=self.xants,
rawMetric=True)
metVals.append(redCorr)
if run_cross_pols:
metNames.append('meanVijXPol')
metNames.append('redCorrXPol')
meanVijXPol = self.mean_Vij_cross_pol_metrics(xants=self.xants,
rawMetric=True)
redCorrXPol = self.red_corr_cross_pol_metrics(xants=self.xants,
rawMetric=True)
metVals.append(meanVijXPol)
metVals.append(redCorrXPol)
# Save all metrics and zscores
metrics, modzScores = {}, {}
for metric, metName in zip(metVals, metNames):
metrics[metName] = metric
modz = per_antenna_modified_z_scores(metric)
modzScores[metName] = modz
for key in metric:
if metName in self.finalMetrics:
self.finalMetrics[metName][key] = metric[key]
self.finalModzScores[metName][key] = modz[key]
else:
self.finalMetrics[metName] = {key: metric[key]}
self.finalModzScores[metName] = {key: modz[key]}
self.allMetrics.update({self.iter: metrics})
self.allModzScores.update({self.iter: modzScores})
def iterative_antenna_metrics_and_flagging(self, crossCut=5, deadCut=5,
alwaysDeadCut=10,
verbose=False,
run_mean_vij=True,
run_red_corr=True,
run_cross_pols=True):
"""Run all four antenna metrics and stores results in self.
Runs all four metrics:
Two for dead antennas
Two for cross-polarized antennas
Saves the results internally to this this antenna metrics object.
Arguments:
crossCut: Modified z-score cut for most cross-polarized antennas.
Default 5 "sigmas".
deadCut: Modified z-score cut for most likely dead antennas.
Default 5 "sigmas".
alwaysDeadCut: Modified z-score cut for definitely dead antennas.
Default 10 "sigmas".
These are all thrown away at once without waiting
to iteratively throw away only the worst offender.
run_mean_vij: Boolean flag which determines if mean_Vij_metrics is executed.
Default is True
run_red_corr: Boolean flag which determines if red_corr_metrics is executed.
Default is True
run_cross_pols: Boolean flag which determines if mean_Vij_cross_pol_metrics and red_corr_cross_pol_metrics are executed.
Default is True
"""
self.reset_summary_stats()
self.find_totally_dead_ants()
self.crossCut, self.deadCut = crossCut, deadCut
self.alwaysDeadCut = alwaysDeadCut
# Loop over
for n in range(len(self.antpols) * len(self.ants)):
self.iter = n
self._run_all_metrics(run_mean_vij=run_mean_vij,
run_red_corr=run_red_corr,
run_cross_pols=run_cross_pols)
# Mostly likely dead antenna
last_iter = list(self.allModzScores)[-1]
worstDeadCutRatio = None
worstCrossCutRatio = None
if run_mean_vij and run_red_corr:
deadMetrics = average_abs_metrics(self.allModzScores[last_iter]['meanVij'],
self.allModzScores[last_iter]['redCorr'])
worstDeadAnt = max(deadMetrics, key=deadMetrics.get)
worstDeadCutRatio = np.abs(deadMetrics[worstDeadAnt]) / deadCut
else:
if run_mean_vij:
deadMetrics = self.allModzScores[last_iter]['meanVij'].copy()
worstDeadAnt = max(deadMetrics, key=deadMetrics.get)
worstDeadCutRatio = (np.abs(deadMetrics[worstDeadAnt])
/ deadCut)
elif run_red_corr:
deadMetrics = self.allModzScores[last_iter]['redCorr'].copy()
worstDeadAnt = max(deadMetrics, key=deadMetrics.get)
worstDeadCutRatio = (np.abs(deadMetrics[worstDeadAnt])
/ deadCut)
if run_cross_pols:
# Most likely cross-polarized antenna
crossMetrics = average_abs_metrics(self.allModzScores[last_iter]['meanVijXPol'],
self.allModzScores[last_iter]['redCorrXPol'])
worstCrossAnt = max(crossMetrics, key=crossMetrics.get)
worstCrossCutRatio = (np.abs(crossMetrics[worstCrossAnt])
/ crossCut)
# Find the single worst antenna, remove it, log it, and run again
if (worstCrossCutRatio >= worstDeadCutRatio
and worstCrossCutRatio >= 1.0):
for antpol in self.antpols:
self.xants.append((worstCrossAnt[0], antpol))
self.crossedAntsRemoved.append((worstCrossAnt[0], antpol))
self.removalIter[(worstCrossAnt[0], antpol)] = n
if verbose:
print('On iteration', n, 'we flag\t', end='')
print((worstCrossAnt[0], antpol))
elif (worstDeadCutRatio > worstCrossCutRatio
and worstDeadCutRatio > 1.0):
dead_ants = set([worstDeadAnt])
for (ant, metric) in deadMetrics.items():
if metric > alwaysDeadCut:
dead_ants.add(ant)
for dead_ant in dead_ants:
self.xants.append(dead_ant)
self.deadAntsRemoved.append(dead_ant)
self.removalIter[dead_ant] = n
if verbose:
print('On iteration', n, 'we flag', dead_ant)
else:
break
def save_antenna_metrics(self, filename):
"""Output all meta-metrics and cut decisions to HDF5 file.
Saves all cut decisions and meta-metrics in an HDF5 that can be loaded
back into a dictionary using hera_qm.ant_metrics.load_antenna_metrics()
Arguments:
filename: The file into which metrics will be written.
"""
if not hasattr(self, 'xants'):
raise KeyError(('Must run AntennaMetrics.'
'iterative_antenna_metrics_and_flagging() first.'))
out_dict = {'xants': self.xants}
out_dict['crossed_ants'] = self.crossedAntsRemoved
out_dict['dead_ants'] = self.deadAntsRemoved
out_dict['final_metrics'] = self.finalMetrics
out_dict['all_metrics'] = self.allMetrics
out_dict['final_mod_z_scores'] = self.finalModzScores
out_dict['all_mod_z_scores'] = self.allModzScores
out_dict['removal_iteration'] = self.removalIter
out_dict['cross_pol_z_cut'] = self.crossCut
out_dict['dead_ant_z_cut'] = self.deadCut
out_dict['always_dead_ant_z_cut'] = self.alwaysDeadCut
out_dict['datafile_list'] = self.dataFileList
out_dict['reds'] = self.reds
metrics_io.write_metric_file(filename, out_dict)