Literature on implementing a Zero-Inflated Beta likelihood

Questions
16

The assemble_coefficient_vector seems to have an error. Where you wrote:

        coefficients = pm.StudentT(f'β_{feature}', nu=4, mu=0, sd=sigma, shape=(N, 1))

        coefficient_vector.append(coefficients)

    coefficient_vector = theano.tensor.concatenate(coefficient_vector, axis=0)

The coefficients random variable is a 2D array of shape (N, 1). Then when you concatenate, you are doing so on axis=0, which means that the output’s shape will be (M*N, 1). You should change those lines to:

        coefficients = pm.StudentT(f'β_{feature}', nu=4, mu=0, sd=sigma, shape=(N,))

        coefficient_vector.append(coefficients)

    coefficient_vector = theano.tensor.stack(coefficient_vector)

Checkout stack's documentation and you’ll see that this will output an (M, N) shaped tensor.