bindata 0.9.2

A library for sampling correlated binary variates.

bindata

A python replication of the omonymous R library bindata, based on the paper "Generation of correlated artificial binary data.", by Friedrich Leisch, Andreas Weingessel, and Kurt Hornik.

The library fully replicates the existing R-package with the following functions:

• bincorr2commonprob
• check_commonprob (check.commonprob in R)
• commonprob2sigma
• condprob
• ra2ba
• rmvbin
• simul_commonprob (simul.commonprob in R)

Precomputed (via Monte Carlo simulations) SimulVals are also available.

Installation

bindata can be installed with pip as:

pip install bindata

How to

Generate uncorrelated variates

import bindata as bnd

margprob = [0.3, 0.9]

X = bnd.rmvbin(N=100_000, margprob=margprob)

Now let's verify the sample marginals and correlations:

import numpy as np

print(X.mean(0))
print(np.corrcoef(X, rowvar=False))

[0.30102 0.9009 ]
[[ 1.         -0.00101357]
 [-0.00101357  1.        ]]

Generate correlated variates

From a correlation matrix

corr = np.array([[1., -0.25, -0.0625],
                 [-0.25,   1.,  0.25],
                 [-0.0625, 0.25, 1.]])
commonprob = bnd.bincorr2commonprob(margprob=[0.2, 0.5, 0.8], 
                                        bincorr=corr)

X = bnd.rmvbin(margprob=np.diag(commonprob), 
                   commonprob=commonprob, N=100_000)
print(X.mean(0))
print(np.corrcoef(X, rowvar=False))

[0.1996  0.50148 0.80076]
[[ 1.         -0.25552    -0.05713501]
 [-0.25552     1.          0.24412401]
 [-0.05713501  0.24412401  1.        ]]

From a joint probability matrix

commonprob = [[1/2, 1/5, 1/6],
              [1/5, 1/2, 1/6],
              [1/6, 1/6, 1/2]]
X = bnd.rmvbin(N=100_000, commonprob=commonprob)

print(X.mean(0))
print(np.corrcoef(X, rowvar=False))

[0.50076 0.50289 0.49718]
[[ 1.         -0.20195239 -0.33343712]
 [-0.20195239  1.         -0.34203855]
 [-0.33343712 -0.34203855  1.        ]]

For a more comprehensive documentation please consult the documentation.

Acknowledgements

• "Generation of correlated artificial binary data.", by Friedrich Leisch, Andreas Weingessel, and Kurt Hornik.

• Icon created by Freepik - Flaticon

Author

Luca Mingarelli, 2022