summary.sperroresterror calculates mean, standard
deviation, median etc. of the calculated error measures
at the specified level (overall, repetition, or fold).
summary.sperrorestpoolederror does the same with
the pooled error, at the overall or repetition level.
1 2 3
Level at which errors are summarized: 0: overall; 1: repetition; 2: fold
additional arguments (currently ignored)
Let's use an example to explain the
argument. E.g., assume we are using 100-repeated 10-fold
pooled=TRUE (default), the
mean and standard deviation calculated when summarizing
level=0 are calculated across the error
estimates obtained for each of the
100*10 = 1000
pooled=FALSE, mean and standard
deviation are calculated across the
repetitions, using the weighted average of the fold-level
errors to calculate an error value for the entire sample.
This will essentially not affect the mean value but of
course the standard deviation of the error.
pooled=FALSE is not recommended, it is mainly for
testing purposes; when the test sets are small (as in
leave-one-out cross-validation, in the extreme case),
err.pooled=TRUE and examine only the
pooled.error component of its result.
Depending on the level of aggregation, a
data.frame with mean, and at level 0 also standard
deviation, median and IQR of the error measures.