Examples of random() in custom distribution?

[Jack_Caster]

@junpenglao

Actually, you should get a (250x100) array by doing just ppc = pm.sample_posterior_predictive(trace, samples=250) , as the size should be inferred from the shape of the observed - that’s how RV defined using regular distribution in pymc3 does.

Let’s see if I understand. If I run ppc = pm.sample_posterior_predictive(trace), then I get a (trace_length, 1) array. Instead I should get a (trace_length, observed_size) array?

EDIT: I feel like we need to declare dist_shape somehow (which is always defined in the class method). A dirty hack would be generate_samples(scipy.stats.norm.rvs, loc=mu_, scale=sd_, size=size, dist_shape=observed.shape)