Resampling Methods

k-fold

mlpy.kfold(nsamples, sets, rseed=0, indexes=None)

K-fold Resampling Method.

Input

• nsamples - [integer] number of samples
• sets - [integer] number of subsets (= number of tr/ts pairs)
• rseed - [integer] random seed
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of sets tuples: ([training indexes], [test indexes])

mlpy.kfoldS(cl, sets, rseed=0, indexes=None)

Stratified K-fold Resampling Method.

Input

• cl - [list (1 or -1)] class label
• sets - [integer] number of subsets (= number of tr/ts pairs)
• rseed - [integer] random seed
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of sets tuples: ([training indexes], [test indexes])

Monte Carlo

mlpy.montecarlo(nsamples, pairs, sets, rseed=0, indexes=None)

Monte Carlo Resampling Method.

Input

• nsamples - [integer] number of samples
• pairs - [integer] number of tr/ts pairs
• sets - [integer] 1/(fraction of data in test sets)
• rseed - [integer] random seed
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of pairs tuples: ([training indexes], [test indexes])

mlpy.montecarloS(cl, pairs, sets, rseed=0, indexes=None)

Stratified Monte Carlo Resampling Method.

Input

• cl - [list (1 or -1)] class label
• pairs - [integer] number of tr/ts pairs
• sets - [integer] 1/(fraction of data in test sets)
• rseed - [integer] random seed
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of pairs tuples: ([training indexes], [test indexes])

Leave-one-out

mlpy.leaveoneout(nsamples, indexes=None)

Leave-one-out Resampling Method.

Input

• nsamples - [integer] number of samples
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of nsamples tuples: ([training indexes], [test indexes])

All Combinations

mlpy.allcombinations(cl, sets, indexes=None)

All Combinations Resampling Method.

Input

• cl - [list (1 or -1)] class label
• sets - [integer] number of subset
• indexes - [list integer] source indexes (None for [0, nsamples-1])

Output

• idx - list of tuples: ([training indexes], [test indexes])

Manual Resampling

mlpy.manresampling(cl, pairs, trp, trn, tsp, tsn, rseed=0)

Manual Resampling.

Input

• cl - [list (1 or -1)] class label
• pairs - [integer] number of tr/ts pairs
• trp - [integer] number of positive samples in training
• trn - [integer] number of negative samples in training
• tsp - [integer] number of positive samples in test
• tsn - [integer] number of negative samples in test

Output

• idx - list of pairs tuples: ([training indexes], [test indexes])

Resampling File

mlpy.resamplingfile(nsamples, file, sep='t')

Resampling file from file.

Returns a list of tuples: ([training indexes],[test indexes])

Read a file in the form:

[test indexes 'sep'-separated for the first  replicate]
[test indexes 'sep'-separated for the second replicate]
                        .
                        .
                        .
[test indexes 'sep'-separated for the last   replicate]

where indexes must be integers in [0, nsamples-1].

Input

• file - [string] test indexes file
• nsamples - [integer] number of samples

Output

• idx - list of tuples: ([training indexes],[test indexes])