MANM-CS

Data generation module for benchmarking methods for causal structure learning (CSL) from mixed discrete-continuous and nonlinear observational data based upon the mixed additive noise model (MANM). The related paper "MANM-CS: Data Generation for Benchmarking Causal Structure Learning from Mixed Discrete-Continuous and Nonlinear Data" was published at the NeurIPS-21 Workshop "Causal Inference & Machine Learning: Why now?".

Getting started

Installation / Usage via pip

python3 -m pip install manm-cs
python3 -m manm_cs --num_nodes 10 --edge_density 0.5 --num_samples 10000 --discrete_node_ratio 0.5

Installation from Source

Get the code

Start by cloning this repository.

git clone git@github.com:hpi-epic/manm-cs.git
cd manm-cs

Install requirements within venv

Please make sure you have Python 3 installed. We tested the execution of our data generation with Python 3.9. We recommend installing the requirements defined in requirements.txt using virtualenv.

MacOS / Linux

# Install virtualenv
python3 -m pip install --user virtualenv

# Create a new virtual environment
python3 -m venv env

# Activate the virtual environment
source env/bin/activate

Windows

# Install virtualenv
py -m pip install --user virtualenv

# Create a new virtual environment
py -m venv env

# Activate the virtual environment
.\env\Scripts\activate

After the creation of a new virtual environment, we can install the project dependencies defined in setup.cfg for both platforms.

python3 -m pip install .

Execute data generation

You can start the data generation with following command. The generated graph and the dataset are saved as ground_truth.gml and samples.csv in the current working directory. Available parameters for data generation can be seen with python3 -m manm_cs --help.

python3 -m manm_cs \
    --num_nodes 10 \
    --edge_density 0.5 \
    --num_samples 10000 \
    --discrete_node_ratio 0.5

Build Release

python3 -m pip install --upgrade build twine
python3 -m build

# Upload to testPyPi
# use __token__ as username and the pypi token as password
python3 -m twine upload --repository testpypi dist/*

# Upload to PyPi
python3 -m twine upload dist/*

Parameters

License

MIT