I would like to generate a sparse matrix using a custom discrete distribution.

E.g.:

from scipy.sparse import random
from scipy import stats

xk = np.arange(7)
pk = np.array([0.1, 0.2, 0.3, 0.1, 0.1, 0.1, 0.1])
custm = stats.rv_discrete(name='custm', values=(xk, pk))
rvs = custm.rvs

dens = 0.5

S = random(1, 8, density=dens, data_rvs=rvs)
print(S.A)

I would expect to get something like so:

[4, 5, 0, 3, 0, 1, 0, 0]

The above code works just fine for rvs of a continuous distribution (such as rvs = stats.norm().rvs), but when I use a discrete distribution I am met with this error:

  File "Python\Python310\site-packages\scipy\sparse\_construct.py", line 875, in random
    vals = data_rvs(k).astype(dtype, copy=False)
AttributeError: 'int' object has no attribute 'astype'

This error persists even if convert the xk array to floats using np.array(xk,dtype=np.float64).

I have considered some work-arounds such as generating array indices in advance and then randomly picking xk values. A bit messy, but do-able. But I suspect I am not understanding about how sparse.random handles random variates?

CodePudding user response：

The docs on rv_discrete.rvs() say this:

size

Defining number of random variates (Default is 1). Note that size has to be given as keyword, not as positional argument.

Source.

However, scipy.sparse.random() passes the size as a positional argument:

vals = data_rvs(k).astype(dtype, copy=False)

Therefore, the fix is to define a lambda function which accepts a positional argument and calls rvs() with a keyword argument.

S = random(1, 8, density=dens, data_rvs=lambda k: rvs(size=k))

CodePudding user response：

One option is to define custm like this:

custm = stats.rv_discrete(name='custm', values=(xk, pk))()

Note the addition of () at the end of that line. This causes custm to be an instance of rv_discrete_frozen. The first argument of the rvs() method of this instance is size, which is what you want.