causal_networkx.StructuralCausalModel#

class causal_networkx.StructuralCausalModel(exogenous, endogenous)[source]#

Structural Causal Model (SCM) class.

Assumes that all exogenous variables are independent of each other. That is no exogenous variable is a function of other exogenous variables passed in.

This assumes the causal independence mechanism, where all exogenous variables are independent of each other.

Parameters:

exogenous : Dict of functions

The exogenous variables and their functional form passed in as values. This forms a symbolic mapping from exogenous variable names to their distribution. The exogenous variable functions should not have any parameters.

endogenous : Dict of lambda functions

The endogenous variable functions may have parameters.

Examples

>>> import numpy as np
>>> rng = np.random.RandomState()
>>> func_uxy = rng.uniform
>>> func_uz = rng.uniform
>>> func_x = lambda u_xy: 2 * u_xy
>>> func_y = lambda x, u_xy: x
>>> func_z = lambda u_z: u_z**2
>>> scm = StructuralCausalModel(
        exogenous={
            "u_xy": func_uxy,
        },
        endogenous={"x": func_x, "y": func_y},
    )
Attributes:

causal_dependencies : dict

A mapping of each variable and its causal dependencies based on the SCM functions.

var_list : list

The list of variable names in the SCM.

_symbolic_runtime : dict

The mapping from each variable in the SCM to the sampled value of that variable. Used when sampling from the SCM.

Methods

get_causal_graph()

Compute the induced causal diagram.

sample([n, include_latents])

Sample from the SCM.
get_causal_graph()[source]#

Compute the induced causal diagram.

Returns:

G : instance of ADMG

The causal graphical model corresponding to the SCM.

:rtype:py:class:ADMG
sample(n=1000, include_latents=True)[source]#

Sample from the SCM.

Parameters:

n : int, optional

Number of samples to generate, by default 1000.

include_latents : bool, optional

Whether to include latent variables in the returned dataset, by default True.

Returns:

result_df : pd.DataFrame

The sampled dataset.

:rtype:py:class:DataFrame

Examples using causal_networkx.StructuralCausalModel#

An introduction to causal graphs and how to use them

An introduction to causal graphs and how to use them#

Fast Causal Inference (FCI) for causal discovery from observational data

Fast Causal Inference (FCI) for causal discovery from observational data#