def discrepancy( self, ystar, y ):
J = len(y)
d = y - ystar
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
disc = np.zeros( J )
for j in range(J):
h = heavyside( d[j] )
if h > 0.5:
disc[j] = d[j]
return disc
def discrepancy(self, ystar, y):
J = len(y)
d = y - ystar
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
disc = np.zeros(J)
for j in range(J):
h = heavyside(d[j])
if h > 0.5:
disc[j] = d[j]
return disc
def loglikelihood( self, observation_statistics, pseudo_statistics ):
# sh = observation_statistics.shape
# if len(sh) > 1:
# N,J = sh
# elif len(sh) == 0:
# N=1
# J=1
# else:
# N = sh[0]
# J = 1
#
# psh = pseudo_statistics.shape
# if len(psh) > 1:
# S,J = psh
# else:
# S = psh[0]
# J = 1
#
# # if S>1:
# # pdb.set_trace()
# if N > 1:
# assert S == 1,"only allow one greater than the other"
# elif S > 1:
# N=S
# start with the difference d = y - y_star
J = len(pseudo_statistics)
d = pseudo_statistics - observation_statistics
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
loglikelihood = np.zeros( J )
for j in range(J):
h = heavyside( d[j] )
if h > 0.5:
loglikelihood[j] = -d[j]/self.epsilon[j]
if self.cliff_at is not None:
if pseudo_statistics[j] == self.cliff_at[j]:
loglikelihood[j] = -np.inf
#pdb.set_trace()
return np.sum(loglikelihood)
def loglikelihood(self, observation_statistics, pseudo_statistics):
# sh = observation_statistics.shape
# if len(sh) > 1:
# N,J = sh
# elif len(sh) == 0:
# N=1
# J=1
# else:
# N = sh[0]
# J = 1
#
# psh = pseudo_statistics.shape
# if len(psh) > 1:
# S,J = psh
# else:
# S = psh[0]
# J = 1
#
# # if S>1:
# # pdb.set_trace()
# if N > 1:
# assert S == 1,"only allow one greater than the other"
# elif S > 1:
# N=S
# start with the difference d = y - y_star
J = len(pseudo_statistics)
d = pseudo_statistics - observation_statistics
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
loglikelihood = np.zeros(J)
for j in range(J):
h = heavyside(d[j])
if h > 0.5:
loglikelihood[j] = -0.5 * pow(d[j] / self.epsilon[j], 2)
if self.cliff_at is not None:
if pseudo_statistics[j] == self.cliff_at[j]:
loglikelihood[j] = -np.inf
#pdb.set_trace()
return np.sum(loglikelihood)
def loglikelihood_kernel( self, observation_statistics, pseudo_statistics ):
# start with the difference d = y - y_star
J = len(pseudo_statistics)
d = pseudo_statistics - observation_statistics
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
loglikelihood = np.zeros( J )
for j in range(J):
h = heavyside( d[j] )
if h > 0.5:
loglikelihood[j] = -d[j]/self.epsilon[j]
if self.cliff_at is not None:
if pseudo_statistics[j] == self.cliff_at[j]:
loglikelihood[j] = -np.inf
#pdb.set_trace()
return np.sum(loglikelihood)
def loglikelihood_kernel(self, observation_statistics, pseudo_statistics):
# start with the difference d = y - y_star
J = len(pseudo_statistics)
d = pseudo_statistics - observation_statistics
if self.direction == self.up:
d *= -1
N = 1
# assume every observation is outside of tube
loglikelihood = np.zeros(J)
for j in range(J):
h = heavyside(d[j])
if h > 0.5:
loglikelihood[j] = -0.5 * pow(d[j] / self.epsilon[j], 2)
if self.cliff_at is not None:
if pseudo_statistics[j] == self.cliff_at[j]:
loglikelihood[j] = -np.inf
#pdb.set_trace()
return np.sum(loglikelihood)