CCNR 0.2.0
- Initial release.
- Fits a Classifier Chain Network (CCN) for multi-label binary
classification.
- Supports k-fold cross-validation with grid search over q, lambda,
and candidate label orders.
- The order in which labels are chained can be given as a permutation,
or as a function of the training data that is evaluated at the start of
every fit.
conditional_entropy_matrix() returns the conditional
entropy of every label given every other label, and
entropy_label_order() turns those entropies into a chain
order, with methods "cebcc1" to "cebcc4" and
"ebcc" following Jun et al. (2019).- Standard methods for fitted models (
predict,
coef, fitted, residuals,
print, summary) and for cross-validation
results (predict, coef, fitted,
print, summary).
- Built-in scoring functions: Hamming accuracy, exact match ratio,
log-likelihood, micro-F1, macro-F1, and mean AUC.
- BFGS optimizer with Wolfe line search and optional random
restarts.
amphibians: presence of seven amphibian species at 189
Polish water reservoirs, with the site characteristics reduced to 25
principal components. Provides a multilabel data set with rare labels
for examples and illustrations.