I'm having trouble finding quantile functions for well-known probability distributions in Python, do they exist? In particular, is there an inverse normal distribution function? I couldn't find anything in either Numpy or Scipy.
2
-
2stackoverflow.com/q/24695174/625914behzad.nouri– behzad.nouri2015-08-10 01:36:53 +00:00Commented Aug 10, 2015 at 1:36
-
2... and stackoverflow.com/questions/20626994/…Warren Weckesser– Warren Weckesser2015-08-10 02:03:09 +00:00Commented Aug 10, 2015 at 2:03
Add a comment
|
2 Answers
Check the .ppf() method of any distribution class in scipy.stats. This is the equivalent of a quantile function (otherwise named as percent point function or inverse CDF)
An example with the exponential distribution from scipy.stats:
# analysis libs
import scipy
import numpy as np
# plotting libs
import matplotlib as mpl
import matplotlib.pyplot as plt
# Example with the exponential distribution
c = 0
lamb = 2
# Create a frozen exponential distribution instance with specified parameters
exp_obj = scipy.stats.expon(c,1/float(lamb))
x_in = np.linspace(0,1,200) # 200 numbers in [0,1], input for ppf()
y_out = exp_obj.ppf(x_in)
plt.plot(x_in,y_out) # graphically check the results of the inverse CDF
Comments
It seems new but I've found this about numpy and quantile. Maybe you can have a look (not tested)
1 Comment
user2699
That gives the empirical quantiles of a set of observations, rather than the exact quantiles of a theoretical distribution the poster is asking for. It is new to numpy, but gives the same functionality as the function
np.percentile.