def __init__(self, arr):
self.M = arr
self.m, self.n = arr.shape
# print 'm, n : %d, %d' % (self.m, self.n)
self.bound = array([1] * (self.m + 1))
self.bit_tree = BitTree(maxsize=self.m)
# self.delta = np.zeros(self.m, dtype=float)
self.model = []
for i in range(self.m):
ne = sum(1 for x in self.M[i] if x == 1)
pe = min(ne / self.n, 1 - 1e-6)
t = (i, ne, pe)
self.model.append(t)
# print 'get model: '
# pprint(self.model)
self.model = sorted(self.model, key=lambda x: -x[1])
for i in range(len(self.model)):
self.bit_tree.update(i, self.model[i][1])
class BaseSegmentUnit(object):
def __init__(self, arr):
self.M = arr
self.m, self.n = arr.shape
# print 'm, n : %d, %d' % (self.m, self.n)
self.bound = array([1] * (self.m + 1))
self.bit_tree = BitTree(maxsize=self.m)
# self.delta = np.zeros(self.m, dtype=float)
self.model = []
for i in range(self.m):
ne = sum(1 for x in self.M[i] if x == 1)
pe = min(ne / self.n, 1 - 1e-6)
t = (i, ne, pe)
self.model.append(t)
# print 'get model: '
# pprint(self.model)
self.model = sorted(self.model, key=lambda x: -x[1])
for i in range(len(self.model)):
self.bit_tree.update(i, self.model[i][1])
# print 'after sort:'
# pprint(self.model)
# pprint(self.M)
def show(self):
print "获取的部分S[]数组"
print self.M
print "model:"
pprint(self.model)
@property
def l(self):
"""
:return: 当前划分得到几个段内分组
"""
return sum(1 for x in self.bound if x == 1) - 1
@property
def lm(self):
"""
:return: 当前划分得到的Lm
"""
return 2 * self.l * log(self.m, 2) + self.m * log(self.m, 2)
@property
def ld(self):
"""
:return: 当前划分得到的Ld
"""
# print 'pr: %f' % (self.pr, )
return -log(self.pr, 2)
@property
def ll(self):
"""
:return: Ll = Ld + Lm
"""
return self.ld + self.lm
@property
def pr(self):
"""
当前分组得到的Pr
:return: Pr (Ld = -log(Pr, 2))
"""
res = 1.0
# print 'bound: '
# print self.bound
for i, v in enumerate(self.bound):
if v == 0 or i == 0:
continue
a = self.prev_bound(i)
if a is None:
continue
px = self.bit_tree.query(a, i - 1) / ((i - a) * self.n)
"""
print 'n: %d' % (self.n, )
print 'a: %d, i: %d' % (a, i)
print 'sn: %d' % (self.bit_tree.query(a, i-1), )
print 'px : %f' % (px, )
"""
f = lambda x: px ** x[1] * (1 - px) ** (self.n - x[1])
for i in range(a, i):
res *= f(self.model[i])
return res
def prev_bound(self, i):
"""
:param i:
:return: i的前一个边界, 复杂度O(n), 可以考虑采用别的方式进行优化(例如保存起来并维护)
"""
a = max(x[0] for x in enumerate(self.bound) if x[1] and x[0] < i)
return a
def next_bound(self, i):
"""
:param i: 边界位置
:return: 返回i的后一个边界
"""
b = min(x[0] for x in enumerate(self.bound) if x[1] and x[0] > i)
return b
def find(self):
"""
:return: 进行算法步骤, 返回一个值表示最优的Ll, 并将结果保存进self.model, self.bound.
可以考虑将结果用更好的方式进行保存.
"""
raise NotImplementedError()
