def pnsig(self, n=None):
"""
The probability that n events are from the signal source.
:Parameters:
n : int
A number of events <= on_cts
:Returns:
p : float OR float array
If n is provided, the probability that n of the on_cts are from
the signal source is returned; with no arguments, an array of
length (on_cts+1) is returned so that p[n] is the probability
that n counts are from the signal source.
"""
# Make sure we've calculate an array of signal count probabilities.
if self.p_signal is None:
self.p_signal = sigctp(self.on_cts, self.on_intvl,
self.off_cts, self.off_intvl)
if n:
if n<0 or n>self.on_cts:
raise ValueError('Number of source counts must be <= on_cts!')
return self.p_signal[n]
else:
return self.p_signal[:] # return copy so cache can't be altered
def sigmp(self, s):
"""
Marginal posterior density for scalar or vector signal rate(s).
The calculation is exact (to machine precision), and requires a sum
with a number of terms equal to the number of on-source counts. In
many cases with a large number of counts, many of the terms will be
negligible. The marginal likelihood method, sigml, can neglect small
terms and may be preferable to sigmp in these cases; its results
will have to be explicitly normalized to produce a posterior pdf.
:Parameters:
s : float OR float array
Signal rate(s) (counts per unit interval)
:Returns:
siglmp : float OR float array
Natural logarithm of the marginal posterior for the signal rate(s)
"""
# Make sure we've calculate an array of signal count probabilities.
if self.p_signal is None:
self.p_signal = sigctp(self.on_cts, self.on_intvl,
self.off_cts, self.off_intvl)
# Treat array or vector argument cases separately.
try:
s = float(s) # raises TypeError for arrays of length != 1
return psigc(s, self.on_intvl, self.p_signal)
except TypeError:
if len(s.shape) != 1:
raise ValueError('sigll handles only 1-D arrays!')
llvals = zeros_like(s)
for i, sval in enumerate(s):
llvals[i] = psigc(sval, self.on_intvl, self.p_signal)
return llvals
