DOC: Provide a working example of inheriting from pymc.Model

[galenseilis]

Issue with current documentation:

The documentation for pymc.Model seems to give some conceptual gist of how it works, but the example given in the documentation does not work.

There are no imports, and I am not sure where pt is coming from in pt.constant(1). I have thought maybe pt was for "PyTensor" but there is no constant in the namespace of pt resulting from import pytensor as pt. There is a pymc.math.constant, but I am unsure if that is the same thing.

But it is also unclear if the imports are the only issue. Here is some code where I have attempted to address the above namespace issues.

from pymc import *

class CustomModel(Model):
    # 1) override init
    def __init__(self, mean=0, sigma=1, name=''):
        # 2) call super's init first, passing model and name
        # to it name will be prefix for all variables here if
        # no name specified for model there will be no prefix
        super().__init__(name, model)
        # now you are in the context of instance,
        # `modelcontext` will return self you can define
        # variables in several ways note, that all variables
        # will get model's name prefix

        # 3) you can create variables with the register_rv method
        self.register_rv(Normal.dist(mu=mean, sigma=sigma), 'v1', initval=1)
        # this will create variable named like '{name::}v1'
        # and assign attribute 'v1' to instance created
        # variable can be accessed with self.v1 or self['v1']

        # 4) this syntax will also work as we are in the
        # context of instance itself, names are given as usual
        Normal('v2', mu=mean, sigma=sigma)

        # something more complex is allowed, too
        half_cauchy = HalfCauchy('sigma', beta=10, initval=1.)
        Normal('v3', mu=mean, sigma=half_cauchy)

        # Deterministic variables can be used in usual way
        Deterministic('v3_sq', self.v3 ** 2)

        # Potentials too
        Potential('p1', pt.constant(1))

# After defining a class CustomModel you can use it in several
# ways

# I:
#   state the model within a context
with Model() as model:
    CustomModel()
    # arbitrary actions

# II:
#   use new class as entering point in context
with CustomModel() as model:
    Normal('new_normal_var', mu=1, sigma=0)

# III:
#   just get model instance with all that was defined in it
model = CustomModel()

# IV:
#   use many custom models within one context
with Model() as model:
    CustomModel(mean=1, name='first')
    CustomModel(mean=2, name='second')

# variables inside both scopes will be named like `first::*`, `second::*`

Running the above I get an AttributeError that I don't think is related because it occurs at super().__init__(name, model).

Traceback (most recent call last):
  File "/usr/lib/python3.10/idlelib/run.py", line 578, in runcode
    exec(code, self.locals)
  File "/home/galen/pymc_inheritance_question.py", line 42, in <module>
    CustomModel()
  File "/home/galen/.local/lib/python3.10/site-packages/pymc/model.py", line 264, in __call__
    instance.__init__(*args, **kwargs)
  File "/home/galen/pymc_inheritance_question.py", line 10, in __init__
    super().__init__(name, model)
  File "/home/galen/.local/lib/python3.10/site-packages/pymc/model.py", line 588, in __init__
    self.add_coords(coords)
  File "/home/galen/.local/lib/python3.10/site-packages/pymc/model.py", line 1080, in add_coords
    for name, values in coords.items():
AttributeError: 'Model' object has no attribute 'items'

Idea or request for content:

Please provide a minimum working example.

[welcome[bot]]

:tada: Welcome to PyMC! :tada: We're really excited to have your input into the project! :sparkling_heart:
If you haven't done so already, please make sure you check out our Contributing Guidelines and Code of Conduct.

[twiecki]

@galenseilis Thanks for reporting, that definitely should be fixed (pt is import pytensor.tensor as pt). But I'd also like to know if what you're trying to do really requires this?

[galenseilis]

@twiecki Thank you for the clarification on pt.

Goals

Sure, let us start with clarifying what I am trying to do. I have three goals. They're sequentially-dependent in the sense that achieving goal 3 depends on achieving goal 2 which depends on achieving goal 1. However, it would still count for something if I achieved goal 1 alone, or only goal 1 and goal 2.

Goal 1

The first is just to understand how it works, unconditioned on "trying to do something else". Is a working example required to understand how pymc.Model works? Maybe not, but I recommend it.

Goal 2

The second goal is to understand where I might use it. What is the use case for making a PyMC model this way? Often a tool is created for a desired set of use cases, and I would like to understand what was the desired use case for being able to build models this way. Is a working example required to understand the use cases of inheriting from pymc.Model? Maybe not, but I recommend it.

Goal 3

The third goal is assess whether it is suitable for a project I am at the product design stage of. The concept is a Python package for random graphs, random hypergraphs, and random simplicial complexes among other things. PyMC is one of the tools I am considering for providing the Monte Carlo methods. If I go with PyMC, I want to make a decision about my model classes inheriting from pymc.Model exor using the usual context wrapper within my classes' __init__ to store the PyMC model as an attribute. Whether I subclass or use the context wrapper will depend on what I learn from goals 1 and 2. Is a working example required to understand what subclassing from pymc.Model would look like in my project? Maybe not, but I recommend it.

[twiecki]

Maybe https://github.com/pymc-devs/pymc-experimental/blob/main/pymc_experimental/model_builder.py#L30 could help you?

[galenseilis]

Maybe https://github.com/pymc-devs/pymc-experimental/blob/main/pymc_experimental/model_builder.py#L30 could help you?

That looks potentially useful for developing NetMC. Certainly something to weigh against in goal 3. If there are alternatives to pymc.Model I should consider them.

[thodson-usgs]

model_builder.py is interesting, but has that stalled?

It would be nice to know some conventions for instantiating complex models, but when I look to the doc, I always come back to the same broken example.

If there are no desire to fix it, could we/I at least remove the example?

[twiecki]

Didn't know it was broken?