How to specify upper and lower limits when using numpy.random.normal

Question

IOK so I want to be able to pick values from a normal distribution that only ever fall between 0 and 1. In some cases I want to be able to basically just return a completely random distribution, and in other cases I want to return values that fall in the shape of a gaussian.

At the moment I am using the following function:

def blockedgauss(mu,sigma):
    while True:
        numb = random.gauss(mu,sigma)
        if (numb > 0 and numb < 1):
            break
    return numb

It picks a value from a normal distribution, then discards it if it falls outside of the range 0 to 1, but I feel like there must be a better way of doing this.

Answer 1

It sounds like you want a truncated normal distribution. Using scipy, you could use scipy.stats.truncnorm to generate random variates from such a distribution:

import matplotlib.pyplot as plt
import scipy.stats as stats

lower, upper = 3.5, 6
mu, sigma = 5, 0.7
X = stats.truncnorm(
    (lower - mu) / sigma, (upper - mu) / sigma, loc=mu, scale=sigma)
N = stats.norm(loc=mu, scale=sigma)

fig, ax = plt.subplots(2, sharex=True)
ax[0].hist(X.rvs(10000), normed=True)
ax[1].hist(N.rvs(10000), normed=True)
plt.show()

The top figure shows the truncated normal distribution, the lower figure shows the normal distribution with the same mean mu and standard deviation sigma.

Answer 2

I came across this post while searching for a way to return a series of values sampled from a normal distribution truncated between zero and 1 (i.e. probabilities). To help anyone else who has the same problem, I just wanted to note that scipy.stats.truncnorm has the built-in capability ".rvs".

So, if you wanted 100,000 samples with a mean of 0.5 and standard deviation of 0.1:

import scipy.stats
lower = 0
upper = 1
mu = 0.5
sigma = 0.1
N = 100000

samples = scipy.stats.truncnorm.rvs(
          (lower-mu)/sigma,(upper-mu)/sigma,loc=mu,scale=sigma,size=N)

This gives a behavior very similar to numpy.random.normal, but within the bounds desired. Using the built-in will be substantially faster than looping to gather samples, especially for large values of N.

Answer 3

In case anybody wants a solution using numpy only, here is a simple implementation using a normal function and a clip (the MacGyver's approach):

    import numpy as np
    def truncated_normal(mean, stddev, minval, maxval):
        return np.clip(np.random.normal(mean, stddev), minval, maxval)

EDIT: do NOT use this!! this is how you shouldn't do it!! for instance,
a = truncated_normal(np.zeros(10000), 1, -10, 10)
may look like it works, but
b = truncated_normal(np.zeros(10000), 100, -1, 1)
will definitely not draw a truncated normal, as you can see in the following histogram:

Sorry for that, hope nobody got hurt! I guess the lesson is, don't try to emulate MacGyver at coding... Cheers,
Andres

Answer 4

I have made an example script by the following. It shows how to use the APIs to implement the functions we wanted, such as generate samples with known parameters, how to compute CDF, PDF, etc. I also attach an image to show this.

#load libraries   
import scipy.stats as stats

#lower, upper, mu, and sigma are four parameters
lower, upper = 0.5, 1
mu, sigma = 0.6, 0.1

#instantiate an object X using the above four parameters,
X = stats.truncnorm((lower - mu) / sigma, (upper - mu) / sigma, loc=mu, scale=sigma)

#generate 1000 sample data
samples = X.rvs(1000)

#compute the PDF of the sample data
pdf_probs = stats.truncnorm.pdf(samples, (lower-mu)/sigma, (upper-mu)/sigma, mu, sigma)

#compute the CDF of the sample data
cdf_probs = stas.truncnorm.cdf(samples, (lower-mu)/sigma, (upper-mu)/sigma, mu, sigma)

#make a histogram for the samples
plt.hist(samples, bins= 50,normed=True,alpha=0.3,label='histogram');

#plot the PDF curves 
plt.plot(samples[samples.argsort()],pdf_probs[samples.argsort()],linewidth=2.3,label='PDF curve')

#plot CDF curve        
plt.plot(samples[samples.argsort()],cdf_probs[samples.argsort()],linewidth=2.3,label='CDF curve')


#legend
plt.legend(loc='best')