Skip to content

Defining Your Causal Assumptions

The first step you should take to use Mjölnir effectively is to encode your causal assumptions into a DAG. This is done with a networkx DiGraph. Here is an example of a collider \(X_1 \rightarrow X_2 \leftarrow X_3\). 

import networkx as nx

dag = nx.DiGraph()

dag.add_edge('X1', 'X2')
dag.add_edge('X3', 'X2')

Training a Model

If you're familiar with the Scikit-Learn API, this should be familiar. Simply import the class Mjolnir. For the sake of example I am going to ignore the DAG we defined earlier, and generate one randomly from scratch.

from mjolnir import datasets
from mjolnir import Mjolnir

dag, data = datasets.make_dag_regression(n=10)

model = Mjolnir(dag)

model.conformal_fit(data)

Symbolically Analyzing the Model

Mjölnir provides some built-in funtionality for performing symbolic math on the model discovered in the symbolic regression.

For starters, let's display the symbolic expressions learned by model.

print(model._get_sympy_exprs())

If you would like to further analyze the model but you're unfamiliar with SymPy or computer algebra, I recommend looking at the Introductory Tutorial for SymPy to get you started.

Do Calculus

Coming soon.