by Josh Dillon last updated October 17, 2022
This notebook parses and summarizes the output of the file_calibration
notebook to produce a report on per-antenna malfunctions on a daily basis.
Quick links:
import numpy as np
import pandas as pd
import glob
import os
import matplotlib.pyplot as plt
from hera_cal import io, utils
from hera_qm import ant_class
from uvtools.plot import plot_antpos, plot_antclass
%matplotlib inline
from IPython.display import display, HTML
# Parse settings from environment
ANT_CLASS_FOLDER = os.environ.get("ANT_CLASS_FOLDER", "./")
SUM_FILE = os.environ.get("SUM_FILE", None)
OC_SKIP_OUTRIGGERS = os.environ.get("OC_SKIP_OUTRIGGERS", "TRUE").upper() == "TRUE"
for param in ['ANT_CLASS_FOLDER', 'SUM_FILE', 'OC_SKIP_OUTRIGGERS']:
print(f"{param} = '{eval(param)}'")
ANT_CLASS_FOLDER = '/mnt/sn1/2460028' SUM_FILE = '/mnt/sn1/2460028/zen.2460028.42005.sum.uvh5' OC_SKIP_OUTRIGGERS = 'True'
if SUM_FILE is not None:
from astropy.time import Time, TimeDelta
utc = Time(float(SUM_FILE.split('zen.')[-1].split('.sum.uvh5')[0]), format='jd').datetime
print(f'Date: {utc.month}-{utc.day}-{utc.year}')
Date: 3-24-2023
# set thresholds for fraction of the day
overall_thresh = .1
all_zero_thresh = .1
eo_zeros_thresh = .1
cross_pol_thresh = .5
bad_fem_thresh = .1
high_power_thresh = .1
low_power_thresh = .1
low_corr_thresh = .1
bad_shape_thresh = .5
excess_rfi_thresh = .1
chisq_thresh = .25
# Load csvs
csv_files = sorted(glob.glob(os.path.join(ANT_CLASS_FOLDER, '*.ant_class.csv')))
jds = [float(f.split('/')[-1].split('zen.')[-1].split('.sum')[0]) for f in csv_files]
tables = [pd.read_csv(f).dropna(axis=0, how='all') for f in csv_files]
table_cols = tables[0].columns[1::2]
class_cols = tables[0].columns[2::2]
print(f'Found {len(csv_files)} csv files starting with {csv_files[0]}')
Found 1849 csv files starting with /mnt/sn1/2460028/zen.2460028.21313.sum.ant_class.csv
# parse ant_strings
ap_strs = np.array(tables[0]['Antenna'])
ants = sorted(set(int(a[:-1]) for a in ap_strs))
translator = ''.maketrans('e', 'n') | ''.maketrans('n', 'e')
# get node numbers
node_dict = {ant: 'Unknown' for ant in ants}
try:
from hera_mc import cm_hookup
hookup = cm_hookup.get_hookup('default')
for ant_name in hookup:
ant = int("".join(filter(str.isdigit, ant_name)))
if ant in node_dict:
if hookup[ant_name].get_part_from_type('node')['E<ground'] is not None:
node_dict[ant] = int(hookup[ant_name].get_part_from_type('node')['E<ground'][1:])
except:
pass
nodes = sorted(set(node_dict.values()))
def classification_array(col):
class_array = np.vstack([t[col] for t in tables])
class_array[class_array == 'good'] = 1.7
class_array[class_array == 'suspect'] = 1
class_array[class_array == 'bad'] = 0
return class_array.astype(float)
if SUM_FILE is not None:
hd = io.HERADataFastReader(SUM_FILE)
ap_tuples = [(int(ap[:-1]), {'e': 'Jee', 'n': 'Jnn'}[ap[-1]]) for ap in ap_strs]
bad_bools = np.mean(classification_array('Antenna Class') == 0, axis=0) > overall_thresh
bad_aps = [ap_tuples[i] for i in np.arange(len(ap_tuples))[bad_bools]]
suspect_bools = np.mean(classification_array('Antenna Class') == 1, axis=0) > overall_thresh
suspect_aps = [ap_tuples[i] for i in np.arange(len(ap_tuples))[suspect_bools] if ap_tuples[i] not in bad_aps]
good_aps = [ap for ap in ap_tuples if ap not in bad_aps and ap not in suspect_aps]
overall_class = ant_class.AntennaClassification(bad=bad_aps, suspect=suspect_aps, good=good_aps)
autos, _, _ = hd.read(bls=[bl for bl in hd.bls if utils.split_bl(bl)[0] == utils.split_bl(bl)[1]], read_flags=False, read_nsamples=False)
avg_unflagged_auto = {pol: np.mean([autos[bl] for bl in autos if bl[2] == pol and overall_class[utils.split_bl(bl)[0]] != 'bad'], axis=(0, 1)) for pol in ['ee', 'nn']}
--------------------------------------------------------------------------- AxisError Traceback (most recent call last) Cell In [9], line 11 9 overall_class = ant_class.AntennaClassification(bad=bad_aps, suspect=suspect_aps, good=good_aps) 10 autos, _, _ = hd.read(bls=[bl for bl in hd.bls if utils.split_bl(bl)[0] == utils.split_bl(bl)[1]], read_flags=False, read_nsamples=False) ---> 11 avg_unflagged_auto = {pol: np.mean([autos[bl] for bl in autos if bl[2] == pol and overall_class[utils.split_bl(bl)[0]] != 'bad'], axis=(0, 1)) for pol in ['ee', 'nn']} Cell In [9], line 11, in <dictcomp>(.0) 9 overall_class = ant_class.AntennaClassification(bad=bad_aps, suspect=suspect_aps, good=good_aps) 10 autos, _, _ = hd.read(bls=[bl for bl in hd.bls if utils.split_bl(bl)[0] == utils.split_bl(bl)[1]], read_flags=False, read_nsamples=False) ---> 11 avg_unflagged_auto = {pol: np.mean([autos[bl] for bl in autos if bl[2] == pol and overall_class[utils.split_bl(bl)[0]] != 'bad'], axis=(0, 1)) for pol in ['ee', 'nn']} File <__array_function__ internals>:200, in mean(*args, **kwargs) File ~/mambaforge/envs/RTP/lib/python3.10/site-packages/numpy/core/fromnumeric.py:3464, in mean(a, axis, dtype, out, keepdims, where) 3461 else: 3462 return mean(axis=axis, dtype=dtype, out=out, **kwargs) -> 3464 return _methods._mean(a, axis=axis, dtype=dtype, 3465 out=out, **kwargs) File ~/mambaforge/envs/RTP/lib/python3.10/site-packages/numpy/core/_methods.py:169, in _mean(a, axis, dtype, out, keepdims, where) 165 arr = asanyarray(a) 167 is_float16_result = False --> 169 rcount = _count_reduce_items(arr, axis, keepdims=keepdims, where=where) 170 if rcount == 0 if where is True else umr_any(rcount == 0, axis=None): 171 warnings.warn("Mean of empty slice.", RuntimeWarning, stacklevel=2) File ~/mambaforge/envs/RTP/lib/python3.10/site-packages/numpy/core/_methods.py:77, in _count_reduce_items(arr, axis, keepdims, where) 75 items = 1 76 for ax in axis: ---> 77 items *= arr.shape[mu.normalize_axis_index(ax, arr.ndim)] 78 items = nt.intp(items) 79 else: 80 # TODO: Optimize case when `where` is broadcast along a non-reduction 81 # axis and full sum is more excessive than needed. 82 83 # guarded to protect circular imports AxisError: axis 1 is out of bounds for array of dimension 1
def print_issue_summary(bad_ant_strs, title, notes='', plot=False):
'''Print report for list of bad antenna polarizations strings'''
unique_bad_antnums = [int(ap[:-1]) for ap in bad_ant_strs]
display(HTML(f'<h2>{title}: ({len(bad_ant_strs)} antpols across {len(set([ba[:-1] for ba in bad_ant_strs]))} antennas)</h2>'))
if len(notes) > 0:
display(HTML(f'<h4>{notes}</h4>'))
if len(bad_ant_strs) > 0:
print(f'All Bad Antpols: {", ".join(bad_ant_strs)}\n')
for node in nodes:
if np.any([node == node_dict[a] for a in unique_bad_antnums]):
aps = [ap for ap in bad_ant_strs if node_dict[int(ap[:-1])] == node]
whole_ants = [str(wa) for wa in set([int(ap[:-1]) for ap in aps if ap.translate(translator) in bad_ant_strs])]
single_pols = [ap for ap in aps if ap.translate(translator) not in bad_ant_strs]
print(f'Node {node}:')
print(f'\tAntpols ({len(aps)} total): {", ".join(aps)}')
print(f'\tWhole Ants ({len(whole_ants)} total): {", ".join(whole_ants)}')
print(f'\tSingle Pols ({len(single_pols)} total): {", ".join(single_pols)}')
if plot and SUM_FILE is not None:
fig, axes = plt.subplots(1, 2, figsize=(12,4), dpi=70, sharey=True, gridspec_kw={'wspace': 0})
for ax, pol in zip(axes, ['ee', 'nn']):
ax.semilogy(autos.freqs / 1e6, avg_unflagged_auto[pol], 'k--', label='Average\nUnflagged\nAuto')
for ap in aps:
ant = int(ap[:-1]), utils.comply_pol(ap[-1])
auto_bl = utils.join_bl(ant, ant)
if auto_bl[2] == pol:
ax.semilogy(autos.freqs / 1e6, np.mean(autos[auto_bl], axis=0), label=ap)
ax.legend()
ax.set_xlim([40, 299])
ax.set_title(f'{title} on Node {node} ({pol}-antennas)')
ax.set_xlabel('Frequency (MHz)')
axes[0].set_ylabel('Single File Raw Autocorrelation')
plt.tight_layout()
plt.show()
# precompute various helpful quantities
all_slopes = np.vstack([t['Autocorr Slope'] for t in tables])
median_slope = np.median(all_slopes)
bad_slopes = np.vstack([t['Autocorr Slope Class'] for t in tables]) == 'bad'
suspect_slopes = np.vstack([t['Autocorr Slope Class'] for t in tables]) == 'suspect'
bad_shapes = np.vstack([t['Autocorr Shape Class'] for t in tables]) == 'bad'
suspect_shapes = np.vstack([t['Autocorr Shape Class'] for t in tables]) == 'suspect'
all_powers = np.vstack([t['Autocorr Power'] for t in tables])
median_power = np.median(all_powers)
bad_powers = np.vstack([t['Autocorr Power Class'] for t in tables]) == 'bad'
suspect_powers = np.vstack([t['Autocorr Power Class'] for t in tables]) == 'suspect'
bad_rfi = np.vstack([t['RFI in Autos Class'] for t in tables]) == 'bad'
suspect_rfi = np.vstack([t['RFI in Autos Class'] for t in tables]) == 'suspect'
# find all zeros
all_zeros_strs = ap_strs[np.mean(np.vstack([t['Dead? Class'] for t in tables]) == 'bad', axis=0) > all_zero_thresh]
# find even/odd zeros
eo_zeros_strs = ap_strs[np.mean(np.vstack([t['Even/Odd Zeros Class'] for t in tables]) == 'bad', axis=0) > eo_zeros_thresh]
eo_zeros_strs = [ap for ap in eo_zeros_strs if ap not in all_zeros_strs]
# find cross-polarized antennas
cross_pol_strs = ap_strs[np.mean(np.vstack([t['Cross-Polarized Class'] for t in tables]) == 'bad', axis=0) > cross_pol_thresh]
# find FEM power issues: must be low power, high slope, and bad or suspect in power, slope, rfi, and shape
fem_off_prod = (bad_powers + .5 * suspect_powers) * (bad_slopes + .5 * suspect_slopes)
fem_off_prod *= (bad_rfi + .5 * suspect_rfi) * (bad_shapes + .5 * suspect_shapes)
fem_off_strs = ap_strs[np.mean(fem_off_prod * (all_powers < median_power) * (all_slopes > median_slope), axis=0) > .1]
# find high power issues
high_power_strs = ap_strs[np.mean(bad_powers & (all_powers > median_power), axis=0) > high_power_thresh]
# find other low power issues
low_power_strs = ap_strs[np.mean(bad_powers & (all_powers < median_power), axis=0) > low_power_thresh]
low_power_strs = [ap for ap in low_power_strs if ap not in all_zeros_strs and ap not in fem_off_strs]
# find low correlation (but not low power)
low_corr_strs = ap_strs[np.mean(np.vstack([t['Low Correlation Class'] for t in tables]) == 'bad', axis=0) > low_corr_thresh]
low_corr_strs = [ap for ap in low_corr_strs if ap not in (set(low_power_strs) | set(all_zeros_strs) | set(fem_off_strs))]
# find bad bandpasses
bad_bandpass_strs = ap_strs[np.mean(bad_shapes, axis=0) > bad_shape_thresh]
bad_bandpass_strs = [ap for ap in bad_bandpass_strs if ap not in (set(low_power_strs) | set(all_zeros_strs) | set(high_power_strs) | set(fem_off_strs))]
# find antennas with excess RFI
excess_rfi_strs = ap_strs[np.mean(np.vstack([t['RFI in Autos Class'] for t in tables]) == 'bad', axis=0) > excess_rfi_thresh]
excess_rfi_strs = [ap for ap in excess_rfi_strs if ap not in (set(low_power_strs) | set(all_zeros_strs) | set(fem_off_strs) |
set(bad_bandpass_strs) | set(high_power_strs))]
# find antennas with high redcal chi^2
chisq_strs = ap_strs[np.mean(np.vstack([t['Redcal chi^2 Class'] for t in tables]) == 'bad', axis=0) > chisq_thresh]
chisq_strs = [ap for ap in chisq_strs if ap not in (set(bad_bandpass_strs) | set(low_power_strs) | set(excess_rfi_strs) | set(low_corr_strs) |
set(all_zeros_strs) | set(high_power_strs) | set(fem_off_strs) | set(eo_zeros_strs))]
if OC_SKIP_OUTRIGGERS:
chisq_strs = [ap for ap in chisq_strs if int(ap[:-1]) < 320]
# collect all results
to_print = [(all_zeros_strs, 'All-Zeros', 'These antennas have visibilities that are more than half zeros.'),
(eo_zeros_strs, 'Excess Zeros in Either Even or Odd Spectra',
'These antennas are showing evidence of packet loss or X-engine failure.', True),
(cross_pol_strs, 'Cross-Polarized', 'These antennas have their east and north cables swapped.'),
(fem_off_strs, 'Likely FEM Power Issue', 'These antennas have low power, anomolously high slopes, and extra channels identified as RFI.', True),
(high_power_strs, 'High Power', 'These antennas have high median power.', True),
(low_power_strs, 'Other Low Power Issues', 'These antennas have low power, but are not all-zeros and not FEM off.', True),
(low_corr_strs, 'Low Correlation, But Not Low Power', 'These antennas are low correlation, but their autocorrelation power levels look OK.'),
(bad_bandpass_strs, 'Bad Bandpass Shapes, But Not Bad Power',
'These antennas have unusual bandpass shapes, but are not all-zeros, high power, low power, or FEM off.', True),
(excess_rfi_strs, 'Excess RFI', 'These antennas have excess strucutre (identified as possible RFI) in their bandpassed relative to the ' + \
'median antenna, but not low or high power or a bad bandpass.', True),
(chisq_strs, 'Redcal chi^2', 'These antennas have been idenfied as not redundantly calibrating well, even after passing the above checks.')]
def print_high_level_summary():
for tp in sorted(to_print, key=lambda x: len(x[0]), reverse=True):
print(f'{len(tp[0])} antpols (on {len(set([ap[:-1] for ap in tp[0]]))} antennas) frequently flagged for {tp[1]}.')
def print_all_issue_summaries():
for tp in to_print:
print_issue_summary(*tp)
print_high_level_summary()
52 antpols (on 38 antennas) frequently flagged for Redcal chi^2. 41 antpols (on 36 antennas) frequently flagged for Likely FEM Power Issue. 34 antpols (on 17 antennas) frequently flagged for All-Zeros. 20 antpols (on 19 antennas) frequently flagged for Bad Bandpass Shapes, But Not Bad Power. 18 antpols (on 12 antennas) frequently flagged for Other Low Power Issues. 16 antpols (on 10 antennas) frequently flagged for High Power. 11 antpols (on 6 antennas) frequently flagged for Excess Zeros in Either Even or Odd Spectra. 10 antpols (on 5 antennas) frequently flagged for Cross-Polarized. 9 antpols (on 9 antennas) frequently flagged for Low Correlation, But Not Low Power. 7 antpols (on 7 antennas) frequently flagged for Excess RFI.
print_all_issue_summaries()
All Bad Antpols: 92e, 92n, 93e, 93n, 94e, 94n, 147e, 147n, 148e, 148n, 149e, 149n, 150e, 150n, 167e, 167n, 168e, 168n, 169e, 169n, 170e, 170n, 189e, 189n, 190e, 190n, 191e, 191n, 208e, 208n, 209e, 209n, 210e, 210n Node 10: Antpols (6 total): 92e, 92n, 93e, 93n, 94e, 94n Whole Ants (3 total): 92, 93, 94 Single Pols (0 total): Node 15: Antpols (22 total): 147e, 147n, 148e, 148n, 149e, 149n, 150e, 150n, 167e, 167n, 168e, 168n, 169e, 169n, 170e, 170n, 189e, 189n, 190e, 190n, 191e, 191n Whole Ants (11 total): 167, 168, 169, 170, 147, 148, 149, 150, 189, 190, 191 Single Pols (0 total): Node 20: Antpols (6 total): 208e, 208n, 209e, 209n, 210e, 210n Whole Ants (3 total): 208, 209, 210 Single Pols (0 total):
All Bad Antpols: 81e, 81n, 82e, 82n, 83e, 83n, 117e, 117n, 118n, 137e, 137n Node 7: Antpols (11 total): 81e, 81n, 82e, 82n, 83e, 83n, 117e, 117n, 118n, 137e, 137n Whole Ants (5 total): 137, 81, 82, 83, 117 Single Pols (1 total): 118n
--------------------------------------------------------------------------- NameError Traceback (most recent call last) Cell In [25], line 1 ----> 1 print_all_issue_summaries() Cell In [23], line 7, in print_all_issue_summaries() 5 def print_all_issue_summaries(): 6 for tp in to_print: ----> 7 print_issue_summary(*tp) Cell In [10], line 21, in print_issue_summary(bad_ant_strs, title, notes, plot) 19 fig, axes = plt.subplots(1, 2, figsize=(12,4), dpi=70, sharey=True, gridspec_kw={'wspace': 0}) 20 for ax, pol in zip(axes, ['ee', 'nn']): ---> 21 ax.semilogy(autos.freqs / 1e6, avg_unflagged_auto[pol], 'k--', label='Average\nUnflagged\nAuto') 22 for ap in aps: 23 ant = int(ap[:-1]), utils.comply_pol(ap[-1]) NameError: name 'avg_unflagged_auto' is not defined
def classification_plot(col):
class_array = classification_array(col)
plt.figure(figsize=(12, len(ants) / 10), dpi=100)
plt.imshow(class_array.T, aspect='auto', interpolation='none', cmap='RdYlGn', vmin=0, vmax=2,
extent=[jds[0] - np.floor(jds[0]), jds[-1] - np.floor(jds[0]), len(ants), 0])
plt.xlabel(f'JD - {int(jds[0])}')
plt.yticks(ticks=np.arange(.5, len(ants)+.5), labels=[ant for ant in ants], fontsize=6)
plt.ylabel('Antenna Number (East First, Then North)')
plt.gca().tick_params(right=True, top=True, labelright=True, labeltop=True)
plt.tight_layout()
plt.title(f'{col}: Green is "good", Yellow is "suspect", Red is "bad"')
This "big green board" shows the overall (i.e. after redundant calibration) classification of antennas on a per-file basis. This is useful for looking at time-dependent effects across the array. While only antenna numbers are labeled, both polarizations are shown, first East then North going down, above and below the antenna's tick mark.
classification_plot('Antenna Class')
# compute flag fractions for all classifiers and antennas
frac_flagged = []
for col in class_cols[1:]:
class_array = np.vstack([t[col] for t in tables])
class_array[class_array == 'good'] = False
class_array[class_array == 'suspect'] = False
class_array[class_array == 'bad'] = True
frac_flagged.append(np.sum(class_array, axis=0))
def plot_flag_frac_all_classifiers():
ticks = []
for i, col in enumerate(list(class_cols[1:])):
ticks.append(f'{col} ({np.nanmean(np.array(frac_flagged).astype(float)[i]) / len(csv_files):.2%})')
plt.figure(figsize=(8, len(ants) / 10), dpi=100)
plt.imshow(np.array(frac_flagged).astype(float).T, aspect='auto', interpolation='none', cmap='viridis')
plt.xticks(ticks=np.arange(len(list(class_cols[1:]))), labels=ticks, rotation=-45, ha='left')
plt.yticks(ticks=np.arange(.5, len(ap_strs)+.5, 2), labels=[ant for ant in ants], fontsize=6)
plt.ylabel('Antenna Number (East First, Then North)')
plt.gca().tick_params(right=True, labelright=True,)
ax2 = plt.gca().twiny()
ax2.set_xticks(ticks=np.arange(len(list(class_cols[1:]))), labels=ticks, rotation=45, ha='left')
plt.colorbar(ax=plt.gca(), label=f'Number of Files Flagged Out of {len(csv_files)}', aspect=50)
plt.tight_layout()
This plot shows the fraction of files flagged for each reason for each antenna. It's useful for seeing which problems are transitory and which ones are more common. Note that not all flags are independent and in particular redcal chi^2 takes an OR of other classifications as an input. Also note that only antenna numbers are labeled, both polarizations are shown, first East then North going down, above and below the antenna's tick mark.
plot_flag_frac_all_classifiers()
def array_class_plot():
fig, axes = plt.subplots(1, 2, figsize=(14, 6), dpi=100, gridspec_kw={'width_ratios': [2, 1]})
plot_antclass(hd.antpos, overall_class, ax=axes[0], ants=[ant for ant in hd.data_ants if ant < 320], legend=False,
title=f'HERA Core: Overall Flagging Based on {overall_thresh:.1%} Daily Threshold')
plot_antclass(hd.antpos, overall_class, ax=axes[1], ants=[ant for ant in hd.data_ants if ant >= 320], radius=50, title='Outriggers')
Overall classification of antenna-polarizations shown on the array layout. If any antenna is marked bad for any reason more than the threshold (default 10%), it is marked bad here. Likewise, if any antenna is marked suspect for more than 10% of the night (but not bad), it's suspect here.
if SUM_FILE is not None: array_class_plot()