Help on class understanding in pymc code

[Peter Otten]

Robert wrote:

> Hi,
> 
> I follow code example at link:
> 
> https://users.obs.carnegiescience.edu/cburns/ipynbs/PyMC.html
> 
> 
> There is the following code line:
> 
> sampler = pymc.MCMC([alpha,betax,betay,eps,model,tau,z_obs,x_true,y_true])
> 
> 
> I want to know the detail of pymc.MCMC, then I get help content of it
> with:
> 
> /////////////
> help(pymc.MCMC)
> Help on class MCMC in module pymc.MCMC:
> 
> class MCMC(pymc.Model.Sampler)
>  |  This class fits probability models using Markov Chain Monte Carlo.
>  |  Each stochastic variable is assigned a StepMethod object, which makes
>  |  it take a single MCMC step conditional on the rest of the model. These
>  |  step methods are called in turn.
>  |  
>  |    >>> A = MCMC(input, db, verbose=0)
>  |  
> \\\\\\\\\\\\\\\\\\
> 
> 
> help('pymc.Model.Sampler')
> no Python documentation found for 'pymc.Model.Sampler'
> 
> 
> help('pymc.Model')
> Help on class Model in pymc:
> 
> pymc.Model = class Model(pymc.Container.ObjectContainer)
>  |  The base class for all objects that fit probability models. Model is
>  |  initialized with:
>  |  
>  |    >>> A = Model(input, verbose=0)
>  |  
>  |    :Parameters:
>  |      - input : module, list, tuple, dictionary, set, object or nothing.
>  |          Model definition, in terms of Stochastics, Deterministics,
>  |          Potentials and Containers. If nothing, all nodes are collected
>  |          from the base namespace.
>  |  
>  |  Attributes:
>  |    - deterministics
>  |    - stochastics (with observed=False)
>  |    - data (stochastic variables with observed=True)
>  |    - variables
>  |    - potentials
>  |    - containers
>  |    - nodes
>  |    - all_objects
>  |    - status: Not useful for the Model base class, but may be used by
>  |    subclasses.
>  |  
>  |  The following attributes only exist after the appropriate method is
>  |  called:
>  |    - moral_neighbors: The edges of the moralized graph. A dictionary,
>  |    keyed by stochastic variable,
>  |      whose values are sets of stochastic variables. Edges exist between
>  |      the key variable and all variables in the value. Created by method
>  |      _moralize.
>  |    - extended_children: The extended children of self's stochastic
>  |    variables. See the docstring of
>  |      extend_children. This is a dictionary keyed by stochastic
>  |      variable.
>  |    - generations: A list of sets of stochastic variables. The members
>  |    of each element only have parents in
>  |      previous elements. Created by method find_generations.
>  |  
>  |  Methods:
>  |     - sample_model_likelihood(iter): Generate and return iter samples
>  |     of p(data and potentials|model).
>  |       Can be used to generate Bayes' factors.
>  |  
>  |  :SeeAlso: Sampler, MAP, NormalApproximation, weight, Container, graph.
>  |  
>  |  Method resolution order:
>  |      Model
>  |      pymc.Container.ObjectContainer
>  |      pymc.six.NewBase
>  |      pymc.Node.ContainerBase
>  |      __builtin__.object
>  |  
>  |  Methods defined here:
>  |  
>  |  __init__(self, input=None, name=None, verbose=-1)
>  |      Initialize a Model instance.
>  |      
>  |      :Parameters:
>  |        - input : module, list, tuple, dictionary, set, object or
>  |        nothing.
>  |            Model definition, in terms of Stochastics, Deterministics,
>  |            Potentials and Containers. If nothing, all nodes are
>  |            collected from the base namespace.
>  |  
>  |  draw_from_prior(self)
>  |      Sets all variables to random values drawn from joint 'prior',
>  |      meaning contributions of data and potentials to the joint
>  |      distribution are not considered.
>  |  
>  |  get_node(self, node_name)
>  |      Retrieve node with passed name
>  |  
>  |  seed(self)
>  |      Seed new initial values for the stochastics.
>  |  
>  |  ----------------------------------------------------------------------
>  |  Data descriptors defined here:
>  |  
>  |  generations
>  |  
>  |  ----------------------------------------------------------------------
>  |  Data and other attributes defined here:
>  |  
>  |  __slotnames__ = []
>  |  
>  |  register = False
>  |  
>  |  ----------------------------------------------------------------------
>  |  Methods inherited from pymc.Container.ObjectContainer:
>  |  
>  |  replace(self, item, new_container, key)
>  |  
>  |  ----------------------------------------------------------------------
>  |  Data descriptors inherited from pymc.Container.ObjectContainer:
>  |  
>  |  value
>  |      A copy of self, with all variables replaced by their values.
>  |  
>  |  ----------------------------------------------------------------------
>  |  Methods inherited from pymc.Node.ContainerBase:
>  |  
>  |  assimilate(self, new_container)
>  |  
>  |  ----------------------------------------------------------------------
>  |  Data descriptors inherited from pymc.Node.ContainerBase:
>  |  
>  |  __dict__
>  |      dictionary for instance variables (if defined)
>  |  
>  |  __weakref__
>  |      list of weak references to the object (if defined)
>  |  
>  |  logp
>  |      The summed log-probability of all stochastic variables (data
>  |      or otherwise) and factor potentials in self.
>  |  
>  |  ----------------------------------------------------------------------
>  |  Data and other attributes inherited from pymc.Node.ContainerBase:
>  |  
>  |  change_methods = []
>  |  
>  |  containing_classes = []
> ---------
> 
> 
> Now, I have puzzles on the class constructor input parameter:
> [alpha,betax,betay,eps,model,tau,z_obs,x_true,y_true]
> 
> 1. 'class MCMC(pymc.Model.Sampler)' says its inheritance is from
> 'pymc.Model.Sampler'
> 
> 2. When I try to get help on 'pymc.Model.Sampler', it says:
>    'no Python documentation found for 'pymc.Model.Sampler'
> 
> 3. When I continue to catch help on 'pymc.Model.Sampler', I don't see
> content mentions 'Sampler'. This complete help message is shown above.
> 
> So, what is 'pymc.Model.Sampler'?

Unfortunately there is a module pymc.Model and a class pymc.Model.Model, and 
in pymc.__init__.py there is code that overwrites the module with the class. 
Therefore when you write

pymc.Model

you get

pymc.Model.Model

as you can see when you type

>>> import pymc
>>> pymc.Model
<class 'pymc.Model.Model'>

To get around this bad naming use

>>> from pymc.Model import Sampler
>>> help(Sampler)