Specifiying range for log-normal distribution in Python-CodePudding

I am generating a random log-normal distribution using the inbuilt function. Is it possible to specify a range for a given mean and sigma? For instance, I want a random distribution between 0 and 0.5 with mean=0.2, sigma=0.5?

import numpy as np 

r=np.random.lognormal(mean=0.2, sigma=0.5, size=(3,3))
print([r])

CodePudding user response：

Not directly, but you can generate more points and filter those that are outside your parameters. With the values you provided, however, there will be very few such points:

np.random.seed(0)
z = np.random.lognormal(mean=0.2, sigma=0.5, size=10_000)

>>> np.sum(z < 0.5) / len(z)
0.0346

As a side note, please know that the parameters mean and sigma of np.random.lognormal() refer to the underlying normal distribution, not to the mean and std of the log-normal points:

np.random.seed(0)
z = np.random.lognormal(mean=0.2, sigma=0.5, size=1000)
y = np.log(y)

>>> np.mean(z), np.std(z)
(1.3886515119063163, 0.7414709414626542)

>>> np.mean(y), np.std(y)
(0.2018986489272414, 0.5034218384643446)

All that said, if you really want what you asked for, you could do:

shape = 3, 3
zmin, zmax = 0, 0.5

n = np.prod(shape)
zc = np.array([])
while True:
    z = np.random.lognormal(mean=0.2, sigma=0.5, size=n * 100)
    z = z[(zmin <= z) & (z < zmax)]
    z = np.r_[zc, z]
    if len(z) >= n:
        break
r = z[:n].reshape(shape)

>>> r
array([[0.34078793, 0.45366409, 0.40183988],
       [0.43387773, 0.46387512, 0.30535007],
       [0.44248787, 0.32316698, 0.48600577]])

Note, however, that the loop above (which in this case is done on average just once), may run forever if your bounds specify an empty space.