我正在做一个简单的稀疏矩阵求幂,a**16
,使用scipy-0.17。 (注意,而不是单元乘法)。但是,在我的机器上(运行Debian stable和Ubuntu LTS),这比使用for循环或执行像a*a*a*a*a*a*a*a*a*a*a*a*a*a*a*a
这样的傻事慢了十倍。这没有意义,所以我认为我做错了什么,但是什么?Scipy稀疏矩阵求幂:a ** 16比a * a * a * a * a * a * a * a * a * a * a * a * a * a * a * a * a *
import scipy.sparse
from time import time
a=scipy.sparse.rand(2049,2049,.002)
print ("Trying exponentiation (a**16)")
t=time()
x=a**16
print (repr(x))
print ("Exponentiation took %f seconds\n" % (time()-t))
print ("Trying expansion (a*a*a*...*a*a)")
t=time()
y=a*a*a*a*a*a*a*a*a*a*a*a*a*a*a*a
print (repr(y))
print ("Expansion took %f seconds\n" % (time()-t))
print ("Trying a for loop (z=z*a)")
t=time()
z=scipy.sparse.eye(2049)
for i in range(16):
z=z*a
print (repr(z))
print ("Looping took %f seconds\n" % (time()-t))
# Sanity check, all approximately the same answer, right?
assert (abs(x-z)>=1e-9).nnz==0
assert (abs(x-y)>=1e-9).nnz==0
无法重现。指数在我的测试中速度更快。 – user2357112
(另外,你正在打印结果的'__repr__'方法,而不是'repr'esentation。) – user2357112
嗯,我想这很适合某人。你在使用scipy-0.17吗? – hackerb9