Report keys

Here is a list of all keys that you may try to extract from reports, for example with the dot notation like report.acc to obtain all accuracy computations or report.min.roc to obtain all receiver operating characteristic curves under minimum reductions. Not all keys are present in all reports.

Info

This is a reminder that viewing reports based on keys comes in addition to user-defined sensitive attribute dimension names and Progress instance names. For example, use report["male"] or report.male to focus on the namesake sensitive attribute dimension - the dict notation is a supplement in case the names have spaces or special characters.

Quantities

These are raw quantities that are contained in measures or reductions.

Name Role Details
samples count the sample count
positives count the positive predictions
negatives count the negative predictions
tp count the true positive predictions
tn count the true negative predictions
ap count the actual positive labels
an count the actual negative labels
freedom parameter the degrees of freedom
slope parameter the slope of pinball deviation
distribution curve the score distribution
roc curve the receiver operating characteristics curve
top count the number of top scores considered
precision metric the precision score
repr metric the representation in top samples

Measures

These are measures of algorithmic performance.

Name Type & role Details
pr classification measure the positive rate
acc classification measure the accuracy
tpr classification measure the true positive rate
tnr classification measure the true negative rate
gmi classification measure the geometric mean between true positive and true negative rate
ppv classification measure the positive predictive value (precision)
f1 classification measure the F1 score
tar classification measure the true acceptance rate
trr classification measure the true rejection rate
lift classification measure the lift over random
mcc classification measure the Matthews correlation coefficient
kappa classification measure the Cohen's kappa
avgscore recommendation measure the average score
auc recommendation measure the area under curve of the receiver operating characteristics
ndcg recommendation measure the normalized discounted cumulative gain
topndcg recommendation measure the NDCG at top recommendations
tophr recommendation measure the hit ratio of top recommendations
toprec recommendation measure the precision of top recommendations
topf1 recommendation measure the F1 score of top recommendations
nmrr recommendation measure the normalized mean reciprocal rank
nentropy recommendation measure the normalized entropy
mabs regression measure mean absolute error
rmse regression measure root mean square error
r2 regression measure coefficient of determination
pinball regression measure pinball deviation
spearman ranking measure Spearman’s rank correlation
rbo ranking measure the rank-biased overlap
ndrl ranking measure the normalized discounted ranking loss

Reductions

Name Role Details
min reduction the minimum
max reduction the maximum
maxerror reduction the maximum deviation from the ideal value
std reduction the standard deviation
gini reduction the gini coefficient
mean reduction the average
wmean reduction the weighted average
maxdiff reduction the maximum difference
maxbarea reduction the maximum area between curves
maxrel reduction the maximum relative difference
largestmaxdiff reduction the maximum difference from the largest group (the whole population if included)
largestmaxrel reduction the maximum relative difference from the largest group (the whole population if included)
largestmaxbarea reduction the maximum area between curves and the curve of the largest group (the whole population if included)