Must be real number, not FreeRV

Questions
1

I have som trouble with my model in pymc3. My main code looks like this:

with pm.Model() as RAA_Eloss:

    #Priors
    a = Normal('a', mu=10, sigma=30)
    b = Normal('b', mu=10, sigma=30)
    c = Normal('c', mu=1, sigma=10)

    mu = RAA_model(a, b, c, RAA_x, RAA_y, RAA_xerr, RAA_yerr, pp_pt, PPspectra)

    likelihood = Normal('likelihood', mu=mu, tau=1/(RAA_yerr)**2, observed=RAA_y)
    trace = pm.sample(draws=5000, tune=1000, chains=4, return_inferencedata=True)
    az.plot_trace(trace);

And the function RAA_model:

def RAA_model(a, b, c, RAA_x, RAA_y, RAA_xerr, RAA_yerr, pt, pp_spectra):
    intg_res = np.zeros_like(RAA_x)
    pp_fit = interp1d(pt, pp_spectra, fill_value="extrapolate")
    for i, x in enumerate(gala30x):
        # integral DeltaPt from 0 to infinity
        scale_fct = RAA_x / gala30x[-1]
        x = x * scale_fct
        shifted_pt = RAA_x + x
        mean_dpt = mean_ptloss(shifted_pt, b, c)
        alpha = a
        beta = a / mean_dpt
        pdpt = Eloss_func(x, alpha, beta)
        intg_res += scale_fct*gala30w[i]*pdpt*pp_fit(shifted_pt)
    return intg_res/pp_fit(RAA_x)

The problem is in the function Eloss_func(x, alpha, beta):

def Eloss_func(x, alpha, beta):
    return beta**(alpha)*np.exp(-beta*x)*x**(alpha-1)/math.gamma(alpha)

It seem to be with math.gamma(alpha), I get the following error: must be real number, not FreeRV
How can in fix this?

2

Looks like you need an implementation of math.gamma that works symbolically.
You can probably just replace it with theano.tensor.gamma. For PyMC v4 it will be at.gamma.

3

Thank you! That worked!