使用您的示例輸入:
>>> list_list_values = [[3, 6, 7], [5, 7, 11, 25, 99], [8, 45], [4, 7, 8],
[0, 1], [21, 31, 41], [9, 11, 22, 33, 44], [17, 19]]
>>> list_comp_values = [0.7, 9.8, 1.2, 5, 10, 11.7, 6, 0.2]
>>> list_range_values = [1, 3, 5]
首先,一些有心計的發電機:
>>> indices_weights = ((list_list_values[i], list_comp_values[i])
for i in list_range_values)
>>> flat_indices_weights = ((i, weight) for indices, weight in indices_weights
for i in indices)
現在我們將數據傳遞到numpy
。我不知道如何從迭代器生成rec.array
,所以我必須將上面的生成器轉換爲列表。也許有避免的方式...
>>> i_w = numpy.rec.array(list(flat_indices_weights),
dtype=[('i', int), ('weight', float)])
>>> numpy.histogram(i_w['i'], bins=range(0, 100), weights=i_w['weight'])
(array([ 0. , 0. , 0. , 0. , 5. , 9.8, 0. , 14.8, 5. ,
0. , 0. , 9.8, 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 11.7, 0. , 0. , 0. , 9.8, 0. ,
0. , 0. , 0. , 0. , 11.7, 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 11.7, 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. ,
0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 9.8]),
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67,
68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84,
85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]))
我過會兒跟進JoshAdel與幾個我自己的測試。到目前爲止最快的解決方案使用Bago的設置,但用內置的histogram
函數取代了sum_by_group
功能。下面是我得到的數字(更新):
方法1(jterrace):2.65
方法2(用於環路):2.25
方法3(巴戈):1.14
方法4(直方圖):2.82
Method5(3/4組合):1.07
需要注意的是這裏實現的,第一種方法,根據我的測試給出了不正確的結果。我沒有時間弄清楚問題所在。我的測試代碼如下所示;它只是輕輕調整JoshAdel的原始代碼,但爲了方便起見,我在這裏完整地發佈它。 (已更新,以包括Bago的評論,並且有些被取消)
from timeit import Timer
setstr="""import numpy as np
import itertools
import random
Nlists = 1000
total_lists = 5000
outsz = 100
maxsublistsz = 100
# create random list of lists
list_range_values = random.sample(xrange(total_lists),Nlists)
list_list_values = [random.sample(xrange(outsz),np.random.randint(1,maxsublistsz)) for k in xrange(total_lists)]
list_comp_values = list(10*np.random.uniform(size=(total_lists,)))
v = np.zeros((outsz,))
def indices(start, end):
lens = end - start
np.cumsum(lens, out=lens)
i = np.ones(lens[-1], dtype=int)
i[0] = start[0]
i[lens[:-1]] += start[1:]
i[lens[:-1]] -= end[:-1]
np.cumsum(i, out=i)
return i
def sum_by_group(values, groups):
order = np.argsort(groups)
groups = groups[order]
values = values[order]
values.cumsum(out=values)
index = np.ones(len(groups), 'bool')
index[:-1] = groups[1:] != groups[:-1]
values = values[index]
groups = groups[index]
values[1:] = np.diff(values)
return values, groups
"""
setstr_test = setstr + "\nprint_v = True\n"
method1="""
list_list_lens = np.array(map(len, list_list_values))
comp_vals_expanded = np.repeat(list_comp_values, list_list_lens)
list_vals_flat = np.fromiter(itertools.chain.from_iterable(list_list_values),dtype=int)
list_list_starts = np.concatenate(([0], np.cumsum(list_list_lens)[:-1]))
toadd = indices(list_list_starts[list_range_values],(list_list_starts + list_list_lens)[list_range_values])
v[list_vals_flat[toadd]] += comp_vals_expanded[toadd]
"""
method2="""
for k in list_range_values:
v[list_list_values[k]] += list_comp_values[k]
"""
method3="""
llv = [np.fromiter(list_list_values[i], 'int') for i in list_range_values]
lcv = [list_comp_values[i] for i in list_range_values]
counts = map(len, llv)
indices = np.concatenate(llv)
values = np.repeat(lcv, counts)
totals, indices_unique = sum_by_group(values, indices)
v[indices_unique] += totals
"""
method4="""
indices_weights = ((list_list_values[i], list_comp_values[i]) for i in list_range_values)
flat_indices_weights = ((i, weight) for indices, weight in indices_weights for i in indices)
i_w = np.rec.array(list(flat_indices_weights), dtype=[('i', 'i'), ('weight', 'd')])
v += np.histogram(i_w['i'], bins=range(0, outsz + 1), weights=i_w['weight'], new=True)[0]
"""
method5="""
llv = [np.fromiter(list_list_values[i], 'int') for i in list_range_values]
lcv = [list_comp_values[i] for i in list_range_values]
counts = map(len, llv)
indices = np.concatenate(llv)
values = np.repeat(lcv, counts)
v += np.histogram(indices, bins=range(0, outsz + 1), weights=values, new=True)[0]
"""
t1 = Timer(method1,setup=setstr).timeit(100)
print t1
t2 = Timer(method2,setup=setstr).timeit(100)
print t2
t3 = Timer(method3,setup=setstr).timeit(100)
print t3
t4 = Timer(method4,setup=setstr).timeit(100)
print t4
t5 = Timer(method5,setup=setstr).timeit(100)
print t5
exec(setstr_test + method1 + "\nprint v\n")
exec("\nv = np.zeros((outsz,))\n" + method2 + "\nprint v\n")
exec("\nv = np.zeros((outsz,))\n" + method3 + "\nprint v\n")
exec("\nv = np.zeros((outsz,))\n" + method4 + "\nprint v\n")
exec("\nv = np.zeros((outsz,))\n" + method5 + "\nprint v\n")
這是上一個問題在http:// stackoverflow中的重新組合。com/questions/8695806/python-list-comprehension-for-numpy,它將在今天晚些時候被刪除。 – 2012-01-02 13:05:00
爲什麼你想擺脫循環?這對我來說似乎很合適。 – 2012-01-02 14:16:54
我已經想出了你以前的問題(使用'numpy.bincount')的一種方式,但是對於這個問題,如果你需要'list_list_values'的每個集合的任意權重,我認爲沒有辦法 – 2012-01-02 14:19:40