The PERT example in the documentation does not work

[rlouf]

When I execute the following code copy-pasted in the documentation at the last commit on main:

import aeppl
import aesara
import aesara.tensor as at

srng = at.random.RandomStream()

def pert(srng, a, b, c):
    r"""Construct a random variable that is PERT-distributed."""
    alpha = 1 + 4 * (b - a) / (c - a)
    beta = 1 + 4 * (c - b) / (c - a)

    X_rv = srng.beta(alpha, beta)

    z = a + (b - a) * X_rv

    return z

A_rv = srng.uniform(10, 20, name="A")
B_rv = srng.uniform(20, 65, name="B")
C_rv = srng.uniform(65, 100, name="C")
Y_rv = pert(srng, A_rv, B_rv, C_rv)

logprob, (y_vv, a_vv, b_vv, c_vv) = aeppl.joint_logprob(Y_rv, A_rv, B_rv, C_rv)

I get the error:

aeppl.joint_logprob.DensityNotFound: Couldn't derive a log-probability for Elemwise{add,no_inplace}(ValuedVariable.out, Elemwise{mul,no_inplace}.0)

I get similar errors if I simplify the example:

import aeppl
import aesara.tensor as at

srng = at.random.RandomStream(0)

A_rv = srng.uniform(10, 20, name="A")
X_rv = srng.beta(1., 1.)
Z_rv = A_rv - X_rv

logprob, (z_vv, a_vv) = aeppl.joint_logprob(Z_rv, A_rv)

[brandonwillard]

My guess is that it's due to changes introduced by #78.