Skip to main content

statistical causality discovery based on cyclic model

# cyclicmodel

Statistical causal discovery based on cyclic model.
This project is under development.

## Summary

Python package that performs statistical causal discovery under the following condition:

1. there are unobserved common factors
2. two-way causal relationship exists

`cyclicmodel` has been developed based on `bmlingam`, which implemented bayesian mixed LiNGAM.

## Example

```import numpy as np
import pymc3 as pm
import cyclicmodel as cym

# Generate synthetic data,
# which assumes causal relation from x1 to x2
n = 200
x1 = np.random.randn(n)
x2 = x1 + np.random.uniform(low=-0.5, high=0.5, size=n)
xs = np.vstack([x1, x2]).T

# Model settings
hyper_params = cym.define_model.CyclicModelParams(
dist_std_noise='log_normal',
df_indvdl=8.0,
dist_l_cov_21='uniform, -0.9, 0.9',
dist_scale_indvdl='uniform, 0.1, 1.0',
dist_beta_noise='uniform, 0.5, 6.0')

# Generate PyMC3 model
model = cym.define_model.get_pm3_model(xs, hyper_params, verbose=10)

# Run variational inference with PyMC3
with model:
fit = pm.FullRankADVI().fit(n=100000)
trace = fit.sample(1000, include_transformed=True)

# Check the posterior mean of the coefficients
print(np.mean(trace['b_21']))  # from x1 to x2
print(np.mean(trace['b_12']))  # from x2 to x1
```

## Installation

```pip install cyclicmodel
```

## Release history Release notifications | RSS feed

This version 0.0.4 0.0.1

## Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for cyclicmodel, version 0.0.4
Filename, size File type Python version Upload date Hashes
Filename, size cyclicmodel-0.0.4-py3-none-any.whl (5.6 kB) File type Wheel Python version py3 Upload date Hashes
Filename, size cyclicmodel-0.0.4.tar.gz (5.4 kB) File type Source Python version None Upload date Hashes