我正在随机抽样多个矩阵A,每个矩阵计算Ax的概率质量函数(pmf)。 x是随机的,带有+ -1的元素,A是带有+ -1的元素的矩阵。到目前为止我的代码看起来像:
from collections import Counter
import numpy as np
import itertools
def pmf(L):
C = Counter(L)
total = float(sum(C.values()))
for key in C:
C[key]/=total
return C
N = 10
h = 2
n = 2**h
X = np.array(list(itertools.product([-1,1],repeat = n))).T
for _ in xrange(N):
A = (np.random.randint(2, size=(h,n)))*2-1
B = np.dot(A,X)
probs = pmf([tuple(x) for x in B.T.tolist()])
print probs
这给出了例如:
Counter({(0, 0): 0.375, (2, 2): 0.25, (-2, -2): 0.25, (-4, -4): 0.0625, (4, 4): 0.0625})
Counter({(0, 0): 0.25, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-2, -2): 0.125, (0, 4): 0.0625, (-4, 0): 0.0625, (4, 0): 0.0625, (0, -4): 0.0625})
Counter({(-2, 0): 0.1875, (0, 2): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (-2, -4): 0.0625, (4, 2): 0.0625, (2, 4): 0.0625, (-4, -2): 0.0625})
Counter({(0, 2): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (-2, 0): 0.1875, (-4, -2): 0.0625, (-2, -4): 0.0625, (4, 2): 0.0625, (2, 4): 0.0625})
Counter({(0, 0): 0.25, (-2, -2): 0.125, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
Counter({(0, 0): 0.25, (2, -2): 0.125, (2, 2): 0.125, (-2, -2): 0.125, (-2, 2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
Counter({(-2, 0): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (0, 2): 0.1875, (-4, 2): 0.0625, (4, -2): 0.0625, (-2, 4): 0.0625, (2, -4): 0.0625})
Counter({(-2, 0): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (0, 2): 0.1875, (-4, 2): 0.0625, (4, -2): 0.0625, (-2, 4): 0.0625, (2, -4): 0.0625})
Counter({(2, 0): 0.1875, (0, -2): 0.1875, (0, 2): 0.1875, (-2, 0): 0.1875, (-4, -2): 0.0625, (-2, -4): 0.0625, (2, 4): 0.0625, (4, 2): 0.0625})
Counter({(0, 0): 0.25, (-2, -2): 0.125, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
我可以手工收集所有相同的收藏品,并计算每个收藏品的数量。例如,使用上面的输出:
2 Counter({(-2, 0): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (0, 2): 0.1875, (-4, 2): 0.0625, (4, -2): 0.0625, (-2, 4): 0.0625, (2, -4): 0.0625})
2 Counter({(0, 0): 0.25, (-2, -2): 0.125, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
1 Counter({(-2, 0): 0.1875, (0, 2): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (-2, -4): 0.0625, (4, 2): 0.0625, (2, 4): 0.0625, (-4, -2): 0.0625})
1 Counter({(2, 0): 0.1875, (0, -2): 0.1875, (0, 2): 0.1875, (-2, 0): 0.1875, (-4, -2): 0.0625, (-2, -4): 0.0625, (2, 4): 0.0625, (4, 2): 0.0625})
1 Counter({(0, 2): 0.1875, (2, 0): 0.1875, (0, -2): 0.1875, (-2, 0): 0.1875, (-4, -2): 0.0625, (-2, -4): 0.0625, (4, 2): 0.0625, (2, 4): 0.0625})
1 Counter({(0, 0): 0.375, (2, 2): 0.25, (-2, -2): 0.25, (-4, -4): 0.0625, (4, 4): 0.0625})
1 Counter({(0, 0): 0.25, (2, -2): 0.125, (2, 2): 0.125, (-2, -2): 0.125, (-2, 2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
1 Counter({(0, 0): 0.25, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-2, -2): 0.125, (0, 4): 0.0625, (-4, 0): 0.0625, (4, 0): 0.0625, (0, -4): 0.0625})
但是,我想要的是这些相同集合中的每一个都要查看导致它们的矩阵。所以这将是前两组的两个矩阵,其余为1。
这样做的好方法是什么?
答案 0 :(得分:2)
因为你想按值对每一个进行分组,你可以使用键的pmf值创建一个新的dict,并为你拥有的矩阵的列表/值设置值,并且为了帮助你可以使用带有list /的defaultdict组。像这样
from collections import Counter, defaultdict
test = Counter({(0, 0): 0.25, (-2, -2): 0.125, (-2, 2): 0.125, (2, 2): 0.125, (2, -2): 0.125, (-4, 0): 0.0625, (0, 4): 0.0625, (0, -4): 0.0625, (4, 0): 0.0625})
result = defaultdict(list)
for k,v in test.iteritems():
result[v].append(k)
print result
#or more readable
for k,v in result.iteritems():
print k,v
输出
0.25 [(0, 0)]
0.125 [(2, 2), (2, -2), (-2, -2), (-2, 2)]
0.0625 [(-4, 0), (0, 4), (0, -4), (4, 0)]
修改强>
现在,如果你想获得产生特定 pmf-value 的矩阵A
,那么单独从 pmf-value 获取它是不可能的,所以需要通过跟踪产生特定 pmf-value 的每个矩阵来解决这个问题,使用与以前相同的defaultdict方法,并使用 pmf 作为关键并重视列表具有 pmf 的矩阵,如下所示:
from collections import Counter, defaultdict
import numpy as np
import itertools
def pmf(L):
C = Counter(L)
total = float(sum(C.values()))
for key in C:
C[key]/=total
return C
N = 10
h = 2
n = 2**h
X = np.array(list(itertools.product([-1,1],repeat = n))).T
result = defaultdict(list)
for _ in xrange(N):
A = (np.random.randint(2, size=(h,n)))*2-1
B = np.dot(A,X)
probs = pmf([tuple(x) for x in B.T.tolist()])
print probs
result[ frozenset(probs.items()) ].append( A ) #append the one you need
现在这部分frozenset(probs.items())
是因为 Counter 是一个不可用的对象,因为它是可变的,因此不能使用dict键,所以我需要通过转换它使它成为不可变的到其项目的 frozenset 。
有了这个,现在我们所有的矩阵都有一个特定的 pmf
输出
>>> for k,v in result.items():
print k
for A in v:
print A
print ""
print "--------------------"
frozenset({((-4, 2), 0.0625), ((2, 0), 0.1875), ((4, -2), 0.0625), ((0, 2), 0.1875), ((-2, 4), 0.0625), ((0, -2), 0.1875), ((-2, 0), 0.1875), ((2, -4), 0.0625)})
[[ 1 1 -1 1]
[-1 -1 1 1]]
[[ 1 1 -1 -1]
[-1 1 1 1]]
--------------------
frozenset({((2, 0), 0.1875), ((-2, -4), 0.0625), ((2, 4), 0.0625), ((0, 2), 0.1875), ((4, 2), 0.0625), ((0, -2), 0.1875), ((-2, 0), 0.1875), ((-4, -2), 0.0625)})
[[-1 1 1 -1]
[ 1 1 1 -1]]
[[ 1 1 1 -1]
[ 1 -1 1 -1]]
[[-1 1 -1 1]
[-1 1 1 1]]
--------------------
frozenset({((0, 0), 0.25), ((-2, 2), 0.125), ((2, -2), 0.125), ((0, -4), 0.0625), ((0, 4), 0.0625), ((-4, 0), 0.0625), ((2, 2), 0.125), ((-2, -2), 0.125), ((4, 0), 0.0625)})
[[-1 -1 -1 1]
[ 1 1 -1 1]]
[[-1 1 1 -1]
[-1 -1 -1 -1]]
[[-1 -1 -1 -1]
[ 1 -1 1 -1]]
[[ 1 1 -1 1]
[ 1 -1 1 1]]
[[ 1 1 1 1]
[-1 -1 1 1]]
--------------------
>>>