Create helper pm.draw() to take draws from a given variable

[ricardoV94]

eval() should be aliased to something more intuitive like sample() or draw() where appropriate.

We have been using eval in our examples as a substitute to the old random method. This is just the standard Aesara debug feature that exists for any node, and shouldn't be used for more than that.

We can have a function wrapper that compiles a "proper" aesara function and takes a given number of draws. For analogy with V3 it could be named pm.random, but I like the pm.draw name better.

with pm.Model() as m:
  x = pm.Normal("x")

x_draws = pm.draw(x, draws=100)

Adding default updates to RNGs could then more justifiable be done in pymc.aesaraf.compile_pymc which already does this for Simulator variables anyway, and which pm.random/ pm.draw would call. That would be one less thing that Distribution has to be concerned with.

pymc/pymc/aesaraf.py

Line 968 in 75ea2a8

# Set the default update of a NoDistribution RNG so that it is automatically

Originally posted by @ricardoV94 in #5308 (comment)

[danhphan]

Hi @ricardoV94, I would like to work on this task. Could you please give some more detailed suggestions (like which file should we put the pm.draw function wrapper, and is there a similar wrapper function to follow). Thank you.

[ricardoV94]

Hi @ricardoV94, I would like to work on this task. Could you please give some more detailed suggestions (like which file should we put the pm.draw function wrapper, and is there a similar wrapper function to follow). Thank you.

Thanks for offering. I assigned the issue to you.

Good question about where it should go... sampling.py perhaps? This is basically a striped down version of prior_predictive. @fonnesbeck what do you think?

A rough sketch of what it could look like:

def draw(vars, draws=1, mode=None, **kwargs):
  if not isinstance(vars, (list, tuple)):
    vars = [vars]
  f = pm.aesaraf.compile_pymc([], vars, mode=mode, **kwargs)
  return [f() for _ in range(draws)]

Type hints would be a nice plus

[danhphan]

Hi yes, thank you for the suggestion.

I am setting up the dev environment with docker and start reading the sample_prior_predictive function in sampling.py and related files. I will let you know if I face any issue :)