SMC sampling from prior stop working

[aloctavodia]

Before #2984 this used to work

with pm.Model() as m:
    alfa = pm.HalfNormal('alfa', 20)
    p = pm.Beta('p', alfa, 1)
pm.distributions.draw_values([m['p']], size=10)

Also a model like this will fail when sampled using SMC, because SMC is initialized by sampling from the prior.

[ColCarroll]

Shoot. Hoping to land something this week that will fix this and some other shape problems.

…

On Tue, Jun 12, 2018, 7:17 PM Osvaldo Martin ***@***.***> wrote: Before #2984 <#2984> this used to work with pm.Model() as m: alfa = pm.HalfNormal('alfa', 20) p = pm.Beta('p', alfa, 1) pm.distributions.draw_values([m['p']], size=10) Also a model like this will fail when sampled using SMC, because SMC is initialized by sampling from the prior. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3025>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACMHEFFcL9jOT3nu07NeZ3z-eMeQr6p_ks5t8Ev3gaJpZM4UlS_3> .

[aloctavodia]

Great! Thank you!

[ColCarroll]

(got to my computer with the PR on it)

The output is (something like)

[array([0.98216117, 0.99751651, 0.98250691, 0.97701074, 0.87199045,
        0.08888242, 0.93154041, 0.88219556, 0.99708625, 0.81510828])]

now. Let me know if that looks wrong!

[aloctavodia]

That looks like a reasonable output.

[ColCarroll]

Fixed by #2983 (output is similar to above on master).