DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets

W. H. E. Davies, Peter Edwards

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

We introduce a novel instance selection method for combining multiple learned models. This technique results in a single comprehensible model. This is to be contrasted with current methods that typically combine models by voting. The core of the technique, the DAGGER (Disjoint Aggregation using Example Reduction) algorithm selects training example instances which provide evidence for each decision region within each local model. A single model is then learned from the union of these selected examples. We describe experiments on models learned from disjoint training sets which show that DAGGER performs as well as weighted voting on this task and that it extracts examples which are more informative than those that can be selected at random. The experiments were conducted on models learned from disjoint subsets generated with a uniform random distribution. DAGGER is actually designed for use on naturally distributed tasks, with non-random distribution. We discuss how one view of the experimental results suggests that DAGGER should work well on this type of problem.
Original languageEnglish
Title of host publicationInstance Selection and Construction for Data Mining
EditorsHuan Liu, Hiroshi Motoda
PublisherSpringer
Pages319-336
Number of pages18
ISBN (Electronic)978-1-4757-3359-4
ISBN (Print)978-1-4419-4861-8
DOIs
Publication statusPublished - 2001

Publication series

NameThe Springer International Series in Engineering and Computer Science
PublisherSpringer
Volume608
ISSN (Print)0893-3405

Fingerprint

Agglomeration
Experiments

Keywords

  • sampling
  • data-mining
  • distributed learning

Cite this

Davies, W. H. E., & Edwards, P. (2001). DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets. In H. Liu, & H. Motoda (Eds.), Instance Selection and Construction for Data Mining (pp. 319-336). (The Springer International Series in Engineering and Computer Science ; Vol. 608). Springer . https://doi.org/10.1007/978-1-4757-3359-4_18

DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets. / Davies, W. H. E.; Edwards, Peter.

Instance Selection and Construction for Data Mining. ed. / Huan Liu; Hiroshi Motoda. Springer , 2001. p. 319-336 (The Springer International Series in Engineering and Computer Science ; Vol. 608).

Research output: Chapter in Book/Report/Conference proceedingChapter

Davies, WHE & Edwards, P 2001, DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets. in H Liu & H Motoda (eds), Instance Selection and Construction for Data Mining. The Springer International Series in Engineering and Computer Science , vol. 608, Springer , pp. 319-336. https://doi.org/10.1007/978-1-4757-3359-4_18
Davies WHE, Edwards P. DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets. In Liu H, Motoda H, editors, Instance Selection and Construction for Data Mining. Springer . 2001. p. 319-336. (The Springer International Series in Engineering and Computer Science ). https://doi.org/10.1007/978-1-4757-3359-4_18
Davies, W. H. E. ; Edwards, Peter. / DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets. Instance Selection and Construction for Data Mining. editor / Huan Liu ; Hiroshi Motoda. Springer , 2001. pp. 319-336 (The Springer International Series in Engineering and Computer Science ).
@inbook{e0962e94080f4940aad4960c386854d6,
title = "DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets",
abstract = "We introduce a novel instance selection method for combining multiple learned models. This technique results in a single comprehensible model. This is to be contrasted with current methods that typically combine models by voting. The core of the technique, the DAGGER (Disjoint Aggregation using Example Reduction) algorithm selects training example instances which provide evidence for each decision region within each local model. A single model is then learned from the union of these selected examples. We describe experiments on models learned from disjoint training sets which show that DAGGER performs as well as weighted voting on this task and that it extracts examples which are more informative than those that can be selected at random. The experiments were conducted on models learned from disjoint subsets generated with a uniform random distribution. DAGGER is actually designed for use on naturally distributed tasks, with non-random distribution. We discuss how one view of the experimental results suggests that DAGGER should work well on this type of problem.",
keywords = "sampling, data-mining, distributed learning",
author = "Davies, {W. H. E.} and Peter Edwards",
year = "2001",
doi = "10.1007/978-1-4757-3359-4_18",
language = "English",
isbn = "978-1-4419-4861-8",
series = "The Springer International Series in Engineering and Computer Science",
publisher = "Springer",
pages = "319--336",
editor = "Huan Liu and Hiroshi Motoda",
booktitle = "Instance Selection and Construction for Data Mining",

}

TY - CHAP

T1 - DAGGER: Instance Selection for Combining Multiple Models Learnt from Disjoint Subsets

AU - Davies, W. H. E.

AU - Edwards, Peter

PY - 2001

Y1 - 2001

N2 - We introduce a novel instance selection method for combining multiple learned models. This technique results in a single comprehensible model. This is to be contrasted with current methods that typically combine models by voting. The core of the technique, the DAGGER (Disjoint Aggregation using Example Reduction) algorithm selects training example instances which provide evidence for each decision region within each local model. A single model is then learned from the union of these selected examples. We describe experiments on models learned from disjoint training sets which show that DAGGER performs as well as weighted voting on this task and that it extracts examples which are more informative than those that can be selected at random. The experiments were conducted on models learned from disjoint subsets generated with a uniform random distribution. DAGGER is actually designed for use on naturally distributed tasks, with non-random distribution. We discuss how one view of the experimental results suggests that DAGGER should work well on this type of problem.

AB - We introduce a novel instance selection method for combining multiple learned models. This technique results in a single comprehensible model. This is to be contrasted with current methods that typically combine models by voting. The core of the technique, the DAGGER (Disjoint Aggregation using Example Reduction) algorithm selects training example instances which provide evidence for each decision region within each local model. A single model is then learned from the union of these selected examples. We describe experiments on models learned from disjoint training sets which show that DAGGER performs as well as weighted voting on this task and that it extracts examples which are more informative than those that can be selected at random. The experiments were conducted on models learned from disjoint subsets generated with a uniform random distribution. DAGGER is actually designed for use on naturally distributed tasks, with non-random distribution. We discuss how one view of the experimental results suggests that DAGGER should work well on this type of problem.

KW - sampling

KW - data-mining

KW - distributed learning

U2 - 10.1007/978-1-4757-3359-4_18

DO - 10.1007/978-1-4757-3359-4_18

M3 - Chapter

SN - 978-1-4419-4861-8

T3 - The Springer International Series in Engineering and Computer Science

SP - 319

EP - 336

BT - Instance Selection and Construction for Data Mining

A2 - Liu, Huan

A2 - Motoda, Hiroshi

PB - Springer

ER -