class Walker:
"""
A walker
"""
def __init__(self):
pass
def initialise(self):
self.point = Point()
self.point.from_prior()
self.num_scalars = len(self.point.scalars)
# Assign a direction for the walker to travel in
self.direction = -np.log(rng.rand(self.num_scalars))
self.direction /= self.direction.sum()
# Give a hard edge to the distribution
self.edge = -np.Inf*np.ones(self.num_scalars)
# Keep track of which choices were made
self.choices = []
self.keep = []
def advance(self, steps=1000):
self.keep.append(self.point.scalars)
choice = rng.choice(np.arange(0, self.num_scalars),\
p=self.direction)
self.choices.append(choice)
self.edge[choice] = self.point.scalars[choice]
accept = 0
for i in xrange(0, steps):
proposal = copy.deepcopy(self.point)
logH = proposal.perturb()
if logH > 0.:
logH = 0.
if np.all(proposal.scalars >= self.edge) \
and rng.rand() <= np.exp(logH):
self.point = proposal
accept += 1
print('Accepted {a}/{b}. Scalars={s}'\
.format(a=accept, b=steps, s=self.point.scalars))
if accept <= 0.05*steps:
return False
return True
