model_config

Model configuration utilities.

Example configuration for a scalar parameter:

scalar_parameter = {
    "dist": "Normal",
    "kwargs": {
        "mu": 0,
        "sigma": 1,
    },
}

Example configuration of a 1D parameter:

vector_parameter = {
    "dist": "Normal",
    "kwargs": {
        "mu": 0,
        "sigma": 1,
    },
    # dims need to be specified now!
    "dims": "channel",
}

Example configuration of 1D parameter with a hierarchical distribution for only the mu:

hierarchical_parameter = {
    "dist": "Normal",
    "kwargs": {
        # Replace the scalar parameter values with additional distributions
        "mu": {
            "dist": "Normal",
            "kwargs": {
                "mu": 0,
                "sigma": 1,
            },
        },
        # Common sigma for all channels
        "sigma": 1,
    },
    "dims": "channel",
}

Example parameter configuration with a hierarchical non-centered distribution:

hierarchical_non_centered_parameter = {
        "dist": "Normal",
        "kwargs": {
            "mu": {"dist": "HalfNormal", "kwargs": {"sigma": 2},},
            "sigma": {"dist": "HalfNormal", "kwargs": {"sigma": 1},},
        },
        "dims": ("channel"),
        "centered": False,
    }

Example configuration of a 2D parameter:

matrix_parameter = {
    "dist": "Normal",
    "kwargs": {
        "mu": 0,
        "sigma": 1,
    },
    # dims need to be specified now!
    "dims": ("channel", "geo"),
}

Model configuration with all of these variables:

model_config = {
    "alpha": scalar_parameter,
    "beta": vector_parameter,
    "gamma": hierarchical_parameter,
    "delta": matrix_parameter,
}

Creating variables from the configuration:

import pymc as pm

from pymc_marketing.model_config import create_distribution_from_config

coords = {
    "channel": ["A", "B", "C"],
    "geo": ["Region1", "Region2"],
}
with pm.Model(coords=coords) as model:
    alpha = create_distribution_from_config("alpha", model_config)
    beta = create_distribution_from_config("beta", model_config)
    gamma = create_distribution_from_config("gamma", model_config)
    delta = create_distribution_from_config("delta", model_config)

Functions

check_for_deeper_nested_distribution(...)	Check if the parameter configuration contains a deeper nested distribution.
create_dim_handler(desired_dims)	Create a function that maps variable shapes to the desired dims.
create_distribution(name, distribution_name, ...)	Create a PyMC distribution with the specified parameters.
create_distribution_from_config(name, config)	Wrapper around create_distribution that uses a configuration dictionary.
create_hierarchical_non_center(name, ...)	Create a hierarchical non-centered distribution.
create_likelihood_distribution(name, ...)	Create observed variable for the model.
get_distribution(name)	Retrieve a PyMC distribution class by name.
handle_1d(var, dims, desired_dims)	Broadcast a 1D variable to the desired dims.
handle_2d(var, dims, desired_dims)	Broadcast a 2D variable to the desired dims.
handle_nested_distribution(name, param, ...)	Handle a nested distribution configuration.
handle_parameter_configurations(name, param, ...)	Handle the parameter configuration for a variable.
handle_parameter_distributions(name, ...)	Loop over the parameter configurations and handle them.
handle_scalar(var, dims, desired_dims)	Broadcast a scalar to the desired dims.

Exceptions

ModelConfigError(param)	Error for invalid model configuration.
MuAlreadyExistsError(param_config)	Error for when 'mu' is present in the nested 'kwargs'.
NestedDistributionError(param)	Error for when a nested distribution is detected where it is not allowed.
NonCenterInvalidDistributionError(name)	Error for when an invalid distribution is used for non-centered hierarchical distribution.
UnsupportedDistributionError	Error for when an unsupported distribution is used.
UnsupportedShapeError	Error for when the shape of the hierarchical variable is not supported.