Files
eos/libeos/file_reader.py

309 lines
14 KiB
Python

"""
Reading of Amor NeXus data files to extract metadata and event stream.
"""
from typing import BinaryIO, Union
import h5py
import numpy as np
import platform
import logging
import subprocess
from datetime import datetime
from orsopy import fileio
from orsopy.fileio.model_language import SampleModel
from . import const
from .header import Header
from .helpers import extract_walltime
from .event_data_types import AmorGeometry, AmorTiming, AmorEventStream, PACKET_TYPE, EVENT_TYPE, PULSE_TYPE, PC_TYPE
try:
import zoneinfo
except ImportError:
# for python versions < 3.9 try to use the backports version
from backports import zoneinfo
# Time zone used to interpret time strings
AMOR_LOCAL_TIMEZONE = zoneinfo.ZoneInfo(key='Europe/Zurich')
if platform.node().startswith('amor'):
NICOS_CACHE_DIR = '/home/amor/nicosdata/amor/cache/'
GREP = '/usr/bin/grep "%s"'
else:
NICOS_CACHE_DIR = None
class AmorEventData:
"""
Read one amor NeXus datafile and extract relevant header information.
Implements EventDatasetProtocol
"""
fileName: str
first_index: int
last_index: int = -1
EOF: bool = False
max_events: int
owner: fileio.Person
experiment: fileio.Experiment
sample: fileio.Sample
instrument_settings: fileio.InstrumentSettings
geometry: AmorGeometry
timing: AmorTiming
data: AmorEventStream
startTime: np.int64
def __init__(self, fileName:Union[str, h5py.File, BinaryIO], first_index:int=0, max_events:int=1_000_000):
if type(fileName) is str:
self.fileName = fileName
self.hdf = h5py.File(fileName, 'r', swmr=True)
elif type(fileName) is h5py.File:
self.fileName = fileName.filename
self.hdf = fileName
else:
self.fileName = repr(fileName)
self.hdf = h5py.File(fileName, 'r')
self.first_index = first_index
self.max_events = max_events
self.read_header_info()
self.read_instrument_configuration()
self.read_event_stream()
# actions applied to any dataset
self.correct_for_chopper_phases()
self.read_chopper_trigger_stream()
self.extract_walltime()
self.read_proton_current_stream()
if type(fileName) is str:
# close the input file to free memory, only if the file was opened in this object
self.hdf.close()
del(self.hdf)
def _replace_if_missing(self, key, nicos_key, dtype=float):
try:
return dtype(self.hdf[f'/entry1/Amor/{key}'][0])
except(KeyError, IndexError):
if NICOS_CACHE_DIR:
try:
logging.warning(f" using parameter {nicos_key} from nicos cache")
year_date = self.fileDate.strftime('%Y')
value = str(subprocess.getoutput(f'{GREP} {NICOS_CACHE_DIR}nicos-{nicos_key}/{year_date}')).split('\t')[-1]
return dtype(value)
except Exception:
logging.error("Couldn't get value from nicos cache", exc_info=True)
return dtype(0)
else:
logging.warning(f" parameter {key} not found, relpace by zero")
return dtype(0)
def read_header_info(self):
# read general information and first data set
title = self.hdf['entry1/title'][0].decode('utf-8')
proposal_id = self.hdf['entry1/proposal_id'][0].decode('utf-8')
user_name = self.hdf['entry1/user/name'][0].decode('utf-8')
user_affiliation = 'unknown'
user_email = self.hdf['entry1/user/email'][0].decode('utf-8')
user_orcid = None
sampleName = self.hdf['entry1/sample/name'][0].decode('utf-8')
model = self.hdf['entry1/sample/model'][0].decode('utf-8')
instrumentName = 'Amor'
source = self.hdf['entry1/Amor/source/name'][0].decode('utf-8')
sourceProbe = 'neutron'
start_time = self.hdf['entry1/start_time'][0].decode('utf-8')
# extract start time as unix time, adding UTC offset of 1h to time string
start_date = datetime.fromisoformat(start_time)
self.fileDate = start_date.replace(tzinfo=AMOR_LOCAL_TIMEZONE)
self.owner = fileio.Person(
name=user_name,
affiliation=user_affiliation,
contact=user_email,
)
if user_orcid:
self.owner.orcid = user_orcid
self.experiment = fileio.Experiment(
title=title,
instrument=instrumentName,
start_date=start_date,
probe=sourceProbe,
facility=source,
proposalID=proposal_id
)
self.sample = fileio.Sample(
name=sampleName,
model=SampleModel(stack=model),
sample_parameters=None,
)
def read_instrument_configuration(self):
chopperSeparation = float(np.take(self.hdf['entry1/Amor/chopper/pair_separation'], 0))
detectorDistance = float(np.take(self.hdf['entry1/Amor/detector/transformation/distance'], 0))
chopperDetectorDistance = detectorDistance-float(np.take(self.hdf['entry1/Amor/chopper/distance'], 0))
polarizationConfigs = ['unpolarized', 'unpolarized', 'po', 'mo', 'op', 'pp', 'mp', 'om', 'pm', 'mm']
mu = self._replace_if_missing('instrument_control_parameters/mu', 'mu', float)
nu = self._replace_if_missing('instrument_control_parameters/nu', 'nu', float)
kap = self._replace_if_missing('instrument_control_parameters/kappa', 'kappa', float)
kad = self._replace_if_missing('instrument_control_parameters/kappa_offset', 'kad', float)
div = self._replace_if_missing('instrument_control_parameters/div', 'div', float)
ch1TriggerPhase = self._replace_if_missing('chopper/ch1_trigger_phase', 'ch1_trigger_phase', float)
ch2TriggerPhase = self._replace_if_missing('chopper/ch2_trigger_phase', 'ch2_trigger_phase', float)
try:
chopperTriggerTime = (float(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time_zero'][7]) \
-float(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time_zero'][0])) \
/7
chopperTriggerTimeDiff = float(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time_offset'][2])
except (KeyError, IndexError):
logging.debug(' chopper speed and phase taken from .hdf file')
chopperSpeed = self._replace_if_missing('chopper/rotation_speed', 'chopper_phase', float)
chopperPhase = self._replace_if_missing('chopper/phase', 'chopper_phase', float)
tau = 30/chopperSpeed
else:
tau = int(1e-6*chopperTriggerTime/2+0.5)*(1e-3)
chopperTriggerPhase = 180e-9*chopperTriggerTimeDiff/tau
chopperSpeed = 30/tau
chopperPhase = chopperTriggerPhase+ch1TriggerPhase-ch2TriggerPhase
self.geometry = AmorGeometry(mu, nu, kap, kad, div,
chopperSeparation, detectorDistance, chopperDetectorDistance)
self.timing = AmorTiming(ch1TriggerPhase, ch2TriggerPhase, chopperSpeed, chopperPhase, tau)
polarizationConfigLabel = self._replace_if_missing('polarization/configuration/value', 'polarizer_config_label', int)
polarizationConfig = fileio.Polarization(polarizationConfigs[polarizationConfigLabel])
logging.debug(f' polarization configuration: {polarizationConfig} (index {polarizationConfigLabel})')
self.instrument_settings = fileio.InstrumentSettings(
incident_angle = fileio.ValueRange(round(mu+kap+kad-0.5*div, 3),
round(mu+kap+kad+0.5*div, 3),
'deg'),
wavelength = fileio.ValueRange(const.lamdaCut, const.lamdaMax, 'angstrom'),
#polarization = fileio.Polarization.unpolarized,
polarization = fileio.Polarization(polarizationConfig)
)
self.instrument_settings.mu = fileio.Value(
round(mu, 3),
'deg',
comment='sample angle to horizon')
self.instrument_settings.nu = fileio.Value(
round(nu, 3),
'deg',
comment='detector angle to horizon')
self.instrument_settings.div = fileio.Value(
round(div, 3),
'deg',
comment='incoming beam divergence')
self.instrument_settings.kap = fileio.Value(
round(kap, 3),
'deg',
comment='incoming beam inclination')
if abs(kad)>0.02:
self.instrument_settings.kad = fileio.Value(
round(kad, 3),
'deg',
comment='incoming beam angular offset')
def update_header(self, header:Header):
"""
Add dataset information into an existing header.
"""
header.owner = self.owner
header.experiment = self.experiment
header.sample = self.sample
header.measurement_instrument_settings = self.instrument_settings
def read_event_stream(self):
"""
Read the actual event data from file. If file is too large, find event index from packets
that allow splitting of file smaller than self.max_events.
"""
packets = np.recarray(self.hdf['/entry1/Amor/detector/data/event_index'].shape, dtype=PACKET_TYPE)
packets.start_index = self.hdf['/entry1/Amor/detector/data/event_index'][:]
packets.Time = self.hdf['/entry1/Amor/detector/data/event_time_zero'][:]
try:
# packet index that matches first event index
start_packet = int(np.where(packets.start_index==self.first_index)[0][0])
except IndexError:
raise IndexError(f'No event packet found starting at event #{self.first_index}')
packets = packets[start_packet:]
nevts = self.hdf['/entry1/Amor/detector/data/event_time_offset'].shape[0]
if (nevts-self.first_index)>self.max_events:
end_packet = np.where(packets.start_index<=(self.first_index+self.max_events))[0][-1]
self.last_index = packets.start_index[end_packet]-1
packets = packets[:end_packet]
else:
self.last_index = nevts-1
self.EOF = True
nevts = self.last_index+1-self.first_index
# adapte packet to event index relation
packets.start_index -= self.first_index
events = np.recarray(nevts, dtype=EVENT_TYPE)
events.tof = np.array(self.hdf['/entry1/Amor/detector/data/event_time_offset'][self.first_index:self.last_index+1])/1.e9
events.pixelID = self.hdf['/entry1/Amor/detector/data/event_id'][self.first_index:self.last_index+1]
self.data = AmorEventStream(events, packets)
def correct_for_chopper_phases(self):
self.data.events.tof += self.timing.tau*(self.timing.ch1TriggerPhase-self.timing.chopperPhase/2)/180
def read_chopper_trigger_stream(self):
chopper1TriggerTime = np.array(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time_zero'][:-2],
dtype=np.int64)
#self.chopper2TriggerTime = self.chopper1TriggerTime + np.array(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time'][:-2], dtype=np.int64)
# + np.array(self.hdf['entry1/Amor/chopper/ch2_trigger/event_time_offset'][:], dtype=np.int64)
if np.shape(chopper1TriggerTime)[0] > 2:
startTime = chopper1TriggerTime[0]
pulseTimeS = chopper1TriggerTime
else:
logging.warn(' no chopper trigger data available, using event steram instead')
startTime = np.array(self.hdf['/entry1/Amor/detector/data/event_time_zero'][0], dtype=np.int64)
stopTime = np.array(self.hdf['/entry1/Amor/detector/data/event_time_zero'][-2], dtype=np.int64)
pulseTimeS = np.arange(startTime, stopTime, self.timing.tau*1e9, dtype=np.int64)
pulses = np.recarray(pulseTimeS.shape, dtype=PULSE_TYPE)
pulses.time = pulseTimeS
# apply filter in case the events were filtered
if self.first_index>0 or not self.EOF:
pulses = pulses[(pulses.time>=self.data.packets.Time[0])&(pulses.time<=self.data.packets.Time[-1])]
self.data.pulses = pulses
self.eventStartTime = startTime
def extract_walltime(self):
# TODO: fix numba type definition after refactor
self.data.events.wallTime = extract_walltime(self.data.events.tof,
self.data.packets.start_index.astype(np.uint64),
self.data.packets.Time)
def read_proton_current_stream(self):
proton_current = np.recarray(self.hdf['entry1/Amor/detector/proton_current/time'].shape, dtype=PC_TYPE)
proton_current.time = self.hdf['entry1/Amor/detector/proton_current/time'][:]
proton_current.current = self.hdf['entry1/Amor/detector/proton_current/value'][:,0]
if self.first_index>0 or not self.EOF:
proton_current = proton_current[(proton_current.time>=self.data.packets.Time[0])&
(proton_current.time<=self.data.packets.Time[-1])]
self.data.proton_current = proton_current
def info(self):
output = ""
for key in ['owner', 'experiment', 'sample', 'instrument_settings']:
value = repr(getattr(self, key)).replace("\n","\n ")
output += f'\n{key}={value},'
output += '\n'
return output
def __repr__(self):
output = (f"AmorEventData({self.fileName!r}) # {self.data.events.shape[0]} events, "
f"{self.data.pulses.shape[0]} pulses")
return output