Skip to contents

Paired Subsampling

Source: R/ResamplingPairedSubsampling.R
mlr_resamplings_paired_subsampling.Rd

Paired Subsampling to enable inference on the generalization error.

Details

The first repeats_in iterations are a standard ResamplingSubsampling and should be used to obtain a point estimate of the generalization error. The remaining iterations should be used to estimate the standard error. Here, the data is divided repeats_out times into two equally sized disjunct subsets, to each of which subsampling which, a subsampling with repeats_in repetitions is applied. See the $unflatten(iter) method to map the iterations to this nested structure.

Point Estimation

When calling $aggregate() on a resample result obtained using this resampling method, only the first repeats_out iterations will be used. See section "Point Estimation" of MeasureCiConZ.

Parameters

  • repeats_in :: integer(1)
    The inner repetitions.

  • repeats_out :: integer(1)
    The outer repetitions.

  • ratio :: numeric(1)
    The proportion of data to use for training.

References

Nadeau, Claude, Bengio, Yoshua (1999). “Inference for the generalization error.” Advances in neural information processing systems, 12.

Super class

mlr3::Resampling -> ResamplingPairedSubsampling

Active bindings

iters

(integer(1))
The total number of resampling iterations.

Methods

Public methods

Inherited methods

Method new()

Creates a new instance of this R6 class.

Usage

ResamplingPairedSubsampling$new()

Method unflatten()

Unflatten the resampling iteration into a more informative representation:

  • inner: The subsampling iteration

  • outer: NA for the first repeats_in iterations. Otherwise it indicates the outer iteration of the paired subsamplings.

  • partition: NA for the first repeats_in iterations. Otherwise it indicates whether the subsampling is applied to the first or second partition Of the two disjoint halfs.

Usage

ResamplingPairedSubsampling$unflatten(iter)

Arguments

iter

(integer(1))
Resampling iteration.

Returns

list(outer, partition, inner)

Method clone()

The objects of this class are cloneable with this method.

Usage

ResamplingPairedSubsampling$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

pw_subs = rsmp("paired_subsampling")
pw_subs
#> 
#> ── <ResamplingPairedSubsampling> : Paired Subsampling ──────────────────────────
#> • Iterations: 315
#> • Instantiated: FALSE
#> • Parameters: repeats_in=15, repeats_out=10, ratio=0.9