sequential: Sequential Algorithm

In AlineTalhouk/splendid: SuPervised Learning ENsemble for Diagnostic IDentification

Sequential Algorithm

Description

Sequentially train top ranked algorithms on each class ordered by class performance and predict a given class using the sequentially trained fits.

Usage

sequential_train(sm, data, class, boxplot = FALSE)

sequential_pred(fit, sm, data, class, boxplot = FALSE)

Arguments

sm	a splendid_model object
data	data frame with rows as samples, columns as features
class	true/reference class vector used for supervised learning
boxplot	if TRUE, boxplots are plotted showing the distribution of F1-scores per class, for every algorithm.
fit	list of fitted models from sequential_train

Details

sequential_train sequentially trains One-Vs-All models until all classes have been classified. Hence for n classes, there are n - 1 sequential fits. sequential_pred predicts class membership for each One-Vs-All sequential model. Performance is evaluated on by-class F1-scores, since these are better for evaluation than other metrics such as accuracy, precision, and recall.

Value

sequential_train returns a list of fits over the top-ranked sequence.

sequential_pred returns a list of two elements

• prob: predicted sequential probabilities

• cm: confusion matrices for each class

Author(s)

Dustin Johnson, Derek Chiu

Examples

dat <- iris[, 1:4]
class <- iris$Species
sm <- splendid_model(dat, class, n = 2, algorithms = c("slda", "xgboost"))
st <- sequential_train(sm, dat, class)
sp <- sequential_pred(st, sm, dat, class)

AlineTalhouk/splendid documentation built on Feb. 23, 2024, 9:37 p.m.