CODEWITHSUNDEEP
pythonscipyvectorizationnormal-distribution
Coty Huang
Coty Huang

Reputation: 11

Why vectorized normal cumulative function in Python scipy.norm.cdf is different from it scalar version

Why vectorized version of scipy.norm.cdf is different from its scalar version? I have to change to survival function (equals to 1-cdf). Is it a bug or there are common usage of term cdf and sf? My python is 3.9.2 and scipy version scipy 1.7.0.

norm.cdf(1.65, loc = 0, scale = 1)
norm([1.65,1.65],scale=1).cdf(np.zeros(2))
norm([1.65,1.65],scale=1).sf(np.zeros(2))

Output,

0.9505285319663519 
array([0.04947147, 0.04947147])
array([0.95052853, 0.95052853])

Upvotes: 1

Views: 461

Answers (1)

ev-br
ev-br

Reputation: 26090

The equivalence is norm.cdf(x, loc, scale) and norm(loc, scale).cdf(x). This holds for both scalar and vectorized arguments:

In [2]: norm.cdf(1.65, loc=0, scale=1)
Out[2]: 0.9505285319663519

In [3]: norm(loc=0, scale=1).cdf(1.65)
Out[3]: 0.9505285319663519

In [4]: norm.cdf([1.65, 1.65], loc=0, scale=1)
Out[4]: array([0.95052853, 0.95052853])

In [5]: norm(loc=0, scale=1).cdf([1.65, 1.65])
Out[5]: array([0.95052853, 0.95052853])

To avoid confusion, I'd recommend to always use loc and scale as keyword arguments, even though the distributions accept them as positionals.

Upvotes: 4

Related Questions