robstat

Robust statistical methods for multivariate data.

This package contains a range of functions that provides robust location and scale estimates of directional and high-dimensional data. Applications for the latter include the median of complex quantities, such as radio interferometric visibilities.

NumPy computations and SciPy minimizations are accelerated with the JAX machine learning library.

R and certain R functions are used in this package, and are run in Python using the rpy2 R-Python bridge.

Installation

Preferred installation method is pip install . in top-level directory. Alternatively, one can use python setup.py install.

Dependencies

• R to be installed
• rpy2 required and to be installed with pip

Included functions

• Geometric median - also known as the L1-median
• Tukey median - as well as other medians that also employ the notion of location depth, e.g. Oja, Spatial (Tukey, 1975)
• Mardia median - for directional data (Mardia, 1972)

Associated publications:

• HERA Memorandum #106: Non-Gaussian Effects and Robust Location Estimates of Aggregated Calibrated Visibilities
• HERA Memorandum #110: Multivariate Outlier Detection Using Robust Mahalanobis Distances

Updated versions can be found in the publications directory.