| best.models {StabPerf} | R Documentation |
Scores the sampled feature-selectors and feature-selector/classifier-combination to find the best model
best.models(fold, checkpointdir=file.path(getOption("edaprakt.tmpdir"), "checkpoints"), n=Inf)
fold |
Numeric. The fold that is assessed. Use cv=0 if the full dataset shall be analysed. |
checkpointdir |
Character. The directory with the calculated checkpoints that are used for assessment. |
n |
The number of best combinations model to return. |
All feature-checkpoints are read and the feature stability score is calculated and divided by log2 of the median length of the signatures (fss). The selector/classifier checkpoints are read to compute the median and mad of the sampled confusion matrices' accuracies. The final score is derived as follows: (1+mad)*(1-median)*(1-fss). The scores are ordered increasing. A comprehensive statistic is saved in ${tmpdir}/scores/fold_${fold}_scores.txt
ranking |
List. A list with the entries scores, selectors and classifiers ordered increasing (best first) by scores. The selectors and classifiers are indices of the global list. |
best.model, feature.stability.score, accuracy
ranking <- best.models(fold=0, n=3)