Parameter#

class pymc_marketing.terms.Parameter(name, prior=<factory>)[source]#

Named free parameter (scalar or dimensional via prior.dims).

Parameter is the general-purpose leaf for free random variables. Intercept is a convenience alias that defaults name to "intercept"; both share the same lifecycle.

Each parameter in the same pm.Model must have a unique name. Two parameters with the same name (even in different expression trees, e.g., a mu expression and a sigma expression) will clash when build_param creates the variables.

Parameters:

namestr: Name for the PyMC variable (required).
priorVariableFactory: Prior distribution for the parameter. Any VariableFactory (Prior, Censored, Scaled, custom) is accepted. When the prior carries dims (e.g. Prior(..., dims="cohort")), get_coords will extract those coordinates from the dataset.

Examples

from pymc_extras.prior import Prior
from pymc_marketing.terms import (
    Parameter,
    Dot,
    Intercept,
    collect_coords,
    register_data,
    build_param,
)

# scalar
mu = Parameter("mu") + Dot(var_name="x", prior=Prior("Normal", dims="feature"))

# dimensional (cohort)
alpha = Parameter("alpha_scale", prior=Prior("HalfNormal", dims="cohort"))

# same concept, GLM sugar
intercept = Intercept("intercept")  # defaults to name="intercept"

Methods

`Parameter.__init__`(name[, prior])
`Parameter.add_coords`(ds)	Add coordinates to the model during `register_data`.
`Parameter.create_variable`()	Build a free parameter variable.
`Parameter.from_dict`(data)	Reconstruct a parameter from its serialized form.
`Parameter.get_coords`(ds)	Collect coordinates from `prior.dims` present in the dataset.
`Parameter.register_data`(ds)	Register shared data variables as `pmd.Data`.
`Parameter.set_data`(ds[, model])	Update shared data variables for out-of-sample prediction.
`Parameter.to_dict`([_orig])

Attributes

`Parameter.name`
`Parameter.prior`