def clustersort(time, value, deadradius):
"""
Given a series of filtered events, remove all points which are close to a
point with higher value, unless the latter has already been removed by an
even higher one.
Parameters
----------
time, value : array (nevents, npoints)
The filter output.
deadradius : scalar
The temporal distance to consider points as close.
Return
------
time, value : array (N,)
N <= nevents * npoints. The value array is sorted, with the time
array matching the correct points in the value array. All events are
merged together.
"""
indices1, length = clusterargsort.clusterargsort(value, time, deadradius)
indices0 = np.repeat(np.arange(len(value)), np.diff(length))
value = value[indices0, indices1]
time = time[indices0, indices1]
idx = np.argsort(value)
return time[idx], value[idx]
def _merge_candidates(self):
"""
Take the output of filters, apply the dead radius, merge events,
sort by filter output value, mark candidates considered true signal.
"""
# dictionaries layout: filter name -> (photon series -> 1D array)
self.times = collections.defaultdict(dict)
self.values = collections.defaultdict(dict)
self.signal = collections.defaultdict(dict)
for k, fhits in self.filtd.items():
for fname in fhits.dtype.names:
time = fhits[fname]['time']
value = fhits[fname]['value']
indices1, length = clusterargsort.clusterargsort(
value, time, self.deadradius)
indices0 = np.repeat(np.arange(len(value)), np.diff(length))
value = value[indices0, indices1]
time = time[indices0, indices1]
timedelta = np.abs(time - self.s1loc)
minindices = argmin_sliced(timedelta, length)
signal = np.zeros(len(time), bool)
signal[minindices] = True
# this will count signals even in dcr, just don't use them
idx = np.argsort(value)
self.times[fname][k] = time[idx]
self.values[fname][k] = value[idx]
self.signal[fname][k] = signal[idx]
