Repository navigation
Tags
Tags
surd.neural: specific mutual information from a classifier surd.neural.surd(Y, nbins) estimates i(t; S_A) with one masked network over all subsets of sources, scored on held-out folds, and assigns the terms with surd.assign. On the systems of E02 it reaches with 1e4 samples the accuracy of the histogram with 1e6, and it handles five or six sources. Needs PyTorch through the extra 'neural'. Tests, docs, README section with the comparison against the histogram, CI job with CPU PyTorch. Version 1.3.0.
Docs, README, CITATION.cff and CHANGELOG for 1.2.0 Sphinx documentation in docs/ with the API of surd and surd.states, adapted from the SURD-states docs. README: installation extras, a section on causality by states with a usage example, the functions and the examples, both BibTeX entries and an acknowledgement to Enrico Foglia for the packaging and documentation of SURD-states. CITATION.cff lists the Communications Physics paper under references.