Pass size to HalfNormal random method

[ColCarroll]

This fixes #3685 -- it was just propagating the size argument to draw_values. I'll do a quick check of other RVs to make sure this is taken care of elsewhere too.

[ColCarroll]

Yep. All other uses include point= and size=, and HalfNormal was just special. I decided not to make that a required argument because Transforms use draw_value without a size argument.

[ColCarroll]

Also, this does not fix the Bound case:

with pm.Model() as model:
    P = pm.Uniform('P', 1, 100)
    sigma_e = pm.Deterministic('sigma_e', 1/P)
    e = pm.Bound(pm.Normal, 0, 1)('e', mu=0, sigma=sigma_e)
    prior = pm.sample_prior_predictive(100_000)

    
values = pm.distributions.draw_values([P, e], size=100_000)

both prior and values are drawn independently. I'll keep thinking about that, but right now I am in favor of merging this so @adrn has a workaround, and I can open a separate issue.

[adrn]

Wow, fast turnaround :)

Thanks @ColCarroll, I'll give this a try tomorrow!

[codecov]

Codecov Report

Merging #3686 into master will decrease coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3686      +/-   ##
==========================================
- Coverage    89.9%   89.89%   -0.01%     
==========================================
  Files         134      134              
  Lines       20166    20176      +10     
==========================================
+ Hits        18130    18137       +7     
- Misses       2036     2039       +3

Impacted Files	Coverage Δ
pymc3/distributions/continuous.py	80.85% <100%> (ø)	⬆️
pymc3/exceptions.py	84.21% <0%> (-15.79%)	⬇️
pymc3/ode/utils.py	100% <0%> (ø)	⬆️
pymc3/tests/test_ode.py	100% <0%> (ø)	⬆️
pymc3/ode/ode.py	94.05% <0%> (+0.3%)	⬆️

[lucianopaz]

Looks good to me. Thanks @colcarrol! Could you just add a line to the release notes, under maintenance?