causal-magpy 0.1.0

A package for causal discovery and causal inference algorithms

MagPy: Causal Discovery and Effect Estimation Framework

MagPy is a Python framework for causal discovery and effect estimation, aimed at uncovering causal relationships in observational data and estimate the impact of interventinos/counterfactuals.

Project Scope

The primary focus of MagPy is causal discovery, which involves:

1. Learning the structure of causal graphs from observational data
2. Handling mixed data types (continuous, categorical, binary)
3. Implementing constraint-based and score-based algorithms
4. Providing tools for causal inference and intervention analysis

Installation

git clone https://github.com/ergodic-ai/magpy.git
cd magpy
pip install .

or

pip install git+https://github.com/ergodic-ai/magpy.git

Main Packages and Usage

1. AStarSearch

The AStarSearch class implements the A* search algorithm for Bayesian network structure learning.

import pandas as pd
from magpy.search.astar import AStarSearch, bic_score_node

data: pd.DataFrame = ...

astar = AStarSearch(data)
astar.run_scoring(func=bic_score_node, parallel=True)
result = astar.search()

Key features:

• Custom scoring functions
• Parallel processing
• Constraint-based search with super_graph and include_graph

2. Oracles

We are introducing a variety of conditional independence tests - oracles. As an example, the MixedDataOracle class handles conditional independence testing for mixed data types.

from magpy.oracles.mixed import MixedDataOracle

oracle = MixedDataOracle(data, threshold=0.05)
independent = oracle(x, y, Z)

3. PC Skeleton Algorithm

The pc_skeleton function implements the PC algorithm for learning the skeleton of a causal graph.

from magpy.search.pcskeleton import pc_skeleton
from magpy.oracles.mixed import MixedDataOracle
import pandas

data: pandas.DataFrame = ... 

nodes = list(data.columns)
oracle = MixedDataOracle(data, threshold=0.05)
skeleton, sepsets = pc_skeleton(oracle, nodes)

3. Quick Start

For continuous data:

import pandas as pd
from magpy.search.astar import AStarSearch, bic_score_node

data = pd.read_csv('your_data.csv')
astar = AStarSearch(data)
astar.run_scoring(func=bic_score_node, parallel=True)
result = astar.search()


import networkx as nx
import matplotlib.pyplot as plt
G = nx.from_pandas_adjacency(result, create_using=nx.DiGraph)
nx.draw(G, with_labels=True)plt.show()

For categorical data, see the sf.ipynb notebook.