Question

也许是在当天晚些时候，我有一个大脑放屁，但是......

如果y> 0且y <1
y定义明确（不是NAN或未定义）

那么1 / y不应该是＆gt; 1？

我看到1 / y == y。（取倒数是无操作的）。

# rr is a Series of dtype=float64
rr = rr.replace(np.inf, np.nan)
rr = rr.replace(-np.inf, np.nan)
rr = rr.replace(0, 999.9)
print rr.sum()
y = rr[(rr>0) & (rr<1)].copy() # include only those values >0, <1
print "A"
print y.tail()
print "B"
print (1./y).tail()
for i in y:
    assert i>=0 and i<=1
for i in y:
    i = 1/i
    assert i>=1    
for i in (1./y):  # Seems like this look should be the same as the former.
    print i, "GONNA FAIL"
    assert i>=1

2125514.43816 #  rr.sum()  is well defined

A #  y.tail()
0
229994    0.893194
229996    0.997238
229999    0.725193
230000    0.980193
230002    0.819778
Name: rr, dtype: float64

B #  1/y.tail()  ALL THE SAME??!
0
229994    0.893194
229996    0.997238
229999    0.725193
230000    0.980193
230002    0.819778
Name: rr, dtype: float64
0.566025929312 GONNA FAIL
Traceback (most recent call last):
  File "<stdin>", line 22, in <module>
AssertionError

...关于熊猫的事情＆＃39; 1 /系列看起来有点时髦。

结果与y.rtruediv(1)相同。

更新：链接到CSV文件：

import pandas as pd
import numpy as np
import io
import requests
url="https://www.dropbox.com/s/2t03ia7vp1vfx0z/rr.csv?dl=1#"
s=requests.get(url).content
rr=pd.read_csv(io.StringIO(s.decode('utf-8')))
rr = rr[rr.columns[-1]].rename('rr')

print pd.__version__  # 0.19.2
print np.__version__  # 1.13.0

这是full code as a jupyter notebook的链接。

更新：更多logs and code in this folder。

Answer 1

是熊猫的错误！

@RyanStout指出这是在pandas = 0.20.3，numpy = 1.12.1中修复的。升级库也为我修复了它。

互惠似乎被打破了... 1 / y == y（但是y定义明确而不是1）

1 个答案: