Issue with shape of GaussianRandomWalk variable

Questions
4

Right, I think the Weibull distribution does not support getting theano tensor as input, which is weird.

For now, you can cherry pick the logp from pm.Weibull and define a pm.DensityDist:

import pymc3 as pm
import numpy as np
import scipy.stats as st
import theano.tensor as tt
bound=pm.distributions.dist_math.bound
alpha, beta = 1.5, 2.
observed = st.weibull_min(alpha, scale=beta).rvs(30)
t = len(observed)
def Weibull_logp(alpha, beta, value):
    return bound(tt.log(alpha) - tt.log(beta)
                  + (alpha - 1) * tt.log(value / beta)
                  - (value / beta)**alpha,
                     value >= 0, alpha > 0, beta > 0)
with pm.Model():
    alpha = pm.InverseGamma('alpha',alpha = 1.0,beta = 1.0)
    sd    = pm.InverseGamma('sd',alpha = 2.0,beta = 2.0)
    s     = pm.GaussianRandomWalk('s',sd = sd,shape = t)
    x     = pm.DensityDist('x', Weibull_logp, observed=dict(alpha=alpha, beta=sd, value=observed))
    trace = pm.sample()