pycolRxC 0.2.0

A Python implementation of the ecolRxC algorithm

pycolRxC

A Python implementation of the ecolRxC algorithm

Description

This project is an implementation of the ecolRxC algorithm, described[1] and originally implemented in R[2] by Jose M. Pavia and Søren Risbjerg Thomsen.

This implementation makes use of NumPy, SciPy and Numba, and reaches a 100-time speed-up compared to the original R implementation or a Python version without using Numba. This has been benchmarked with a small dataset, the same as in example.py, using timeit for the Python implementation and microbenchmark for the original R library. On this dataset, the R implementation took on average between 3 and 5 minutes, while this implementation takes about 2 to 3 seconds.

For a description of the algorithm, please consult [1], and for a simple example on using this library look at example.py.

Also, please note that only a small subset of the original library's features have been implemented. Notably, only the probit-scale Thomsen method for matrices larger than 2-by-2 has been implemented. In addition, the following have not been implemented:

• census changes

• references

• confidence intervals

• Yule’s approximation

However, most of the groundwork has been laid such that ir should be trivial to add all the capabilities of the original.

[1] Pavía, J. M., & Thomsen, S. R. (2024). ecolRxC: Ecological inference estimation of R × C tables using latent structure approaches. Political Science Research and Methods, 1–19. doi:10.1017/psrm.2024.57

[2] https://cran.r-project.org/web/packages/ecolRxC/index.html

Using pycolRxC from within R

It might be useful to use pycolRxC inside R, as an alternative to the original implementation. This can be accomplished using the reticulate package.

The file example.R contains a simple example on using pycolRxC within R.