使用 Numpy 生成 N 维矩阵

Question

For a certain assignment, I have to create a multivariate discrete probability mass function over N random variables. I want to do this by creating an array A filled with random numbers where each element denotes the joint probability over the random variables. In case of 2 random variables, having i and j possible values respectively, this can be done by creating an (i*j) Numpy array filled with random numbers where the total sum = 1.

It becomes more difficult however, when an additional random variable with k possible values is introduced. In this case, I need to have an i*j*k Numpy array, again filled with random numbers where the total sum equals 1.

Say I am given the structure (number of possible values for each random variable) as a list [n1,n2,...,nN] , how can I from here create such an N dimensional Numpy array?

Answer 1

If l is your list of dimensions, you could let

a = np.random.random(size=l)
a = a/a.sum()

Answer 2

I found the following solution:

def randomArray(structure):
    rand_array = np.random.randint(0, 100, size=(structure))
    my_sum = np.sum(rand_array)
    return rand_array/my_sum

where structure is a list as defined in the question above.