goal: Make two kinds of poisson random variables(λ=30, 60)

condition:

1. We have an array made with the following code(np.random.binomial(n=1, p=1/2, size=1000))
2. If the number in the array is 1, then we make a poisson random variables(λ=30) else we make a poisson random variables(λ=60)

So I wrote the code like this

import numpy as np
a = np.random.binomial(n=1, p=1/2, size=1000)
b = np.where(a==1, np.random.poisson(lam=30), np.random.poisson(lam=60))
print(b)

But the result was like this

[62 62 28 ... 28 28 62]

Fixed random variables were continually being created.

So how can I get the result that random variables are not fixed? (I don't wanna use loop(for or while))

CodePudding user response：

The issue here is that the parameters in np.where are evaluated before it's called, so it's exactly the same as if you write

value1 = np.random.poisson(lam=30)
value2 = np.random.poisson(lam=60)
b = np.where(a==1, value1, value2)

Use a for loop would be a good solution but if you don't want to do that, you can try something with np.vectorize

def poisson(x):
    return np.random.poisson(lam=30) if x == 1 else np.random.poisson(lam=60)

b = np.vectorize(poisson)(a)

should works for example.

CodePudding user response：

@BlackRaven's answer works fine, but the use of np.vectorize uses a Python loop "behind the scenes", so it is not as efficient as it could be. An alternative is to use a to create a vector (i.e. 1-d NumPy array) of lambda values to pass to poisson(), like this (note that I've switched to the newer random interface of NumPy):

rng = np.random.default_rng()
a = rng.binomial(n=1, p=1/2, size=1000)  # array containing 0 and 1
lam = 30   (1-a)*30                      # array containing 30 and 60
b = rng.poisson(lam=lam)                 # mixture of Poisson samples

Of course, you don't have to create a named variable to hold the lambda values:

rng = np.random.default_rng()
a = rng.binomial(n=1, p=1/2, size=1000)  # array containing 0 and 1
b = rng.poisson(lam=30   (1-a)*30)       # mixture of Poisson samples

All the loops involved are now performed within NumPy functions that are implemented in C.