def LaserStates(data_dir, Nt, O_Par):
global Y
aM = Nt * [None]
ll = EKF.LogLike(O_Par, Y[0:Nt], aM=aM)
f = open(join(data_dir, 'LaserStates'), 'w')
for x in aM:
print(x.A[0, 0], x.A[2, 0], file=f)
f.close()
return aM
def NegLogLike(P):
"""Use EKF.LogLike to calculate the log likelihood of the first Nt
values in the Y sequence.
"""
global Y, Nt, best
rad, Tr = P[0:2]
devEp, devEta = P[8:10]
if rad * rad > 0.25 or Tr < 0.5 or devEp < 0.0 or devEta < 0.0:
return 1e20 # If parameters are bad, return big number
try:
L = ParamCalc(*P[:10])
cost = -EKF.LogLike(L, Y[0:Nt]) + (devEta / 1e-2)**2
except (ValueError, RuntimeError) as err:
#print 'NegLogLike returning 1e20 because ', err
return 1e20
if cost < best - 1:
best = cost
print("best=%f" % best)
return cost
def LaserLogLike(
data_dir,
O_Par,
N_step, # Number of steps away from peak in each direction
Nt):
s = O_Par[4]
b = O_Par[5]
Frange = numpy.arange(-N_step, N_step + 1, 1.0)
s_range = (Frange * 5e-3 / N_step + 1.0) * s
b_range = (Frange * 5e-3 / N_step + 1.0) * b
Par = copy.copy(O_Par)
f = open(join(data_dir, 'LaserLogLike'), 'w')
for ss in s_range:
for bb in (b_range):
Par[4] = ss
Par[5] = bb
ll = EKF.LogLike(Par, Y[0:Nt])
print(ss, bb, ll, file=f)
print(' ', file=f)
f.close()
def NLL_Fixed_Noise(P):
"""Use EKF.LogLike to calculate the log likelihood of the first Nnll
values in the Y sequence. Like NegLogLike() but noise is fixed.
"""
global Y, best
Nnll = 250
rad = P[0]
Tr = P[1]
if rad**2 > 0.25 or Tr < 0.5:
return 1e20
try:
L = ParamCalc(*(list(P[0:8]) + [4.0 / P[7], 1e-7]))
# P[7] is scale. The above line sticks dev_eps = 4.0 channels
# and dev_eta = 1e-7 on to the end of the parameter list and
# converts from (rad,Tr) to ic.
LL = EKF.LogLike(L, Y[0:Nnll])
except ValueError as err:
#print 'NLL_Fixed_noise returning 1e20 because ', err
return 1e20
if -LL < best - 1:
best = -LL
print("best=%f" % best)
return (-LL)