SDTEST Estimate classifier error/performance on test set ERR=SDTEST(PD,DATA) % using trained pipeline and data set ERR=SDTEST(DATA,PD) % using trained pipeline and data set ERR=SDTEST(DATA,DEC) % from a test set and decisions ERR=SDTEST(LAB,DEC) % from labels and decisions Specifying performance measures: [ERR,R,T]=SDTEST(DATA,PD,'measures',{'class-errors','TPr','apple','precision','banana'}) INPUTS DATA Test set PD Trained pipeline returning decisions LAB Ground truth labels (SDLAB) DEC Classifier decisions (SDLAB) OUTPUT ERR Estimated error/performance measures (scalar or vector) R SDROC object with performances T Time of execution in seconds DESCRIPTION SDTEST estimates errors or performances by applying trained pipeline PD on the test set or by comparing true labels and decisions. By default, SDTEST returns mean error over classes. Other measures may be defined using 'measures' option similarly to the SDROC. MEASURES TP,FP,TN,FN - true/false positives and negatives (counts of samples) param: target class TPr,FPr,TNr,FNr - fractions of true/false positives/negatives (normalized by the total number of samples per class), param: target class sensitivity - identical to TPr specificity - identical to TNr class-errors - all per-class errors mean-error - mean error over classes (parameter: class priors, default: equal priors) precision - TP/(TP+FP) How much of what we search for, we find among our decisions. (param: target class) posfrac - (TP+FP)/N How much of all observations we flag as positive detrate - detection rate, (TP+FP)/Nt, (param: target class) SEE ALSO SDROC
sdtest
is referenced in examples: