测试运行时间

测试numpy的安装是否优化

#%pylab
import numpy as np
from  datetime  import datetime

n  = 2000
t1 = datetime.now()
U,D,V = np.linalg.svd(np.random.randn(n,n)) 
# U,D,V = np.linalg.svd(np.arange(n*n).reshape(n,n))
dt = datetime.now() - t1
dt.total_seconds()

• 发现jax，在一些算法上很快

In [15]: import jax.numpy as jnp
In [16]: import numpy as np
In [17]: %time u,d,v = jnp.linalg.svd(np.random.randn(2000, 2000))
CPU times: user 3.97 s, sys: 29.1 ms, total: 4 s
Wall time: 759 ms

In [18]: %time u,d,v = np.linalg.svd(np.random.randn(2000, 2000))
CPU times: user 13.4 s, sys: 92.6 ms, total: 13.5 s
Wall time: 2.37 s