In that case, you can evaluate the likelihood of y belonging to each component, and then normalized the likelihood, something like:
ynew = np.asarray([.5])
complogp = y_obs.distribution._comp_logp(theano.shared(ynew))
f_complogp = model_mixed.model.fastfn(complogp)
weight_ynew = []
for ichain in range(trace_mixed.nchains):
for point_idx in range(len(trace_mixed)):
point = trace_mixed._straces[ichain].point(point_idx)
point.pop('sigma')
point.pop('w')
prob = np.exp(f_complogp(point))
prob /= prob.sum()
weight_ynew.append(prob)
weight_ynew = np.asarray(weight_ynew).squeeze()
sns.kdeplot(weight_ynew[:, 0]);