Bagging Equalizes Influence View Full Text


Ontology type: schema:ScholarlyArticle      Open Access: True


Article Info

DATE

2004-06

AUTHORS

Yves Grandvalet

ABSTRACT

Bagging constructs an estimator by averaging predictors trained on bootstrap samples. Bagged estimates almost consistently improve on the original predictor. It is thus important to understand the reasons for this success, and also for the occasional failures. It is widely believed that bagging is effective thanks to the variance reduction stemming from averaging predictors. However, seven years from its introduction, bagging is still not fully understood. This paper provides experimental evidence supporting the hypothesis that bagging stabilizes prediction by equalizing the influence of training examples. This effect is detailed in two different frameworks: estimation on the real line and regression. Bagging’s improvements/deteriorations are explained by the goodness/badness of highly influential examples, in situations where the usual variance reduction argument is at best questionable. Finally, reasons for the equalization effect are advanced. They support that other resampling strategies such as half-sampling should provide qualitatively identical effects while being computationally less demanding than bootstrap sampling. More... »

PAGES

251-270

Identifiers

URI

http://scigraph.springernature.com/pub.10.1023/b:mach.0000027783.34431.42

DOI

http://dx.doi.org/10.1023/b:mach.0000027783.34431.42

DIMENSIONS

https://app.dimensions.ai/details/publication/pub.1013861997


Indexing Status Check whether this publication has been indexed by Scopus and Web Of Science using the SN Indexing Status Tool
Incoming Citations Browse incoming citations for this publication using opencitations.net

JSON-LD is the canonical representation for SciGraph data.

TIP: You can open this SciGraph record using an external JSON-LD service: JSON-LD Playground Google SDTT

[
  {
    "@context": "https://springernature.github.io/scigraph/jsonld/sgcontext.json", 
    "about": [
      {
        "id": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/17", 
        "inDefinedTermSet": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/", 
        "name": "Psychology and Cognitive Sciences", 
        "type": "DefinedTerm"
      }, 
      {
        "id": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/1701", 
        "inDefinedTermSet": "http://purl.org/au-research/vocabulary/anzsrc-for/2008/", 
        "name": "Psychology", 
        "type": "DefinedTerm"
      }
    ], 
    "author": [
      {
        "affiliation": {
          "alternateName": "Heudiasyc, UMR CNRS 6599, Universit\u00e9 de Technologie de Compi\u00e8gne, France", 
          "id": "http://www.grid.ac/institutes/grid.462261.5", 
          "name": [
            "Heudiasyc, UMR CNRS 6599, Universit\u00e9 de Technologie de Compi\u00e8gne, France"
          ], 
          "type": "Organization"
        }, 
        "familyName": "Grandvalet", 
        "givenName": "Yves", 
        "id": "sg:person.015255215731.52", 
        "sameAs": [
          "https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.015255215731.52"
        ], 
        "type": "Person"
      }
    ], 
    "citation": [
      {
        "id": "sg:pub.10.1007/bf00058655", 
        "sameAs": [
          "https://app.dimensions.ai/details/publication/pub.1002929950", 
          "https://doi.org/10.1007/bf00058655"
        ], 
        "type": "CreativeWork"
      }, 
      {
        "id": "sg:pub.10.1023/a:1007515423169", 
        "sameAs": [
          "https://app.dimensions.ai/details/publication/pub.1017116781", 
          "https://doi.org/10.1023/a:1007515423169"
        ], 
        "type": "CreativeWork"
      }, 
      {
        "id": "sg:pub.10.1023/a:1007607513941", 
        "sameAs": [
          "https://app.dimensions.ai/details/publication/pub.1041829946", 
          "https://doi.org/10.1023/a:1007607513941"
        ], 
        "type": "CreativeWork"
      }
    ], 
    "datePublished": "2004-06", 
    "datePublishedReg": "2004-06-01", 
    "description": "Bagging constructs an estimator by averaging predictors trained on bootstrap samples. Bagged estimates almost consistently improve on the original predictor. It is thus important to understand the reasons for this success, and also for the occasional failures. It is widely believed that bagging is effective thanks to the variance reduction stemming from averaging predictors. However, seven years from its introduction, bagging is still not fully understood. This paper provides experimental evidence supporting the hypothesis that bagging stabilizes prediction by equalizing the influence of training examples. This effect is detailed in two different frameworks: estimation on the real line and regression. Bagging\u2019s improvements/deteriorations are explained by the goodness/badness of highly influential examples, in situations where the usual variance reduction argument is at best questionable. Finally, reasons for the equalization effect are advanced. They support that other resampling strategies such as half-sampling should provide qualitatively identical effects while being computationally less demanding than bootstrap sampling.", 
    "genre": "article", 
    "id": "sg:pub.10.1023/b:mach.0000027783.34431.42", 
    "inLanguage": "en", 
    "isAccessibleForFree": true, 
    "isPartOf": [
      {
        "id": "sg:journal.1125588", 
        "issn": [
          "0885-6125", 
          "1573-0565"
        ], 
        "name": "Machine Learning", 
        "publisher": "Springer Nature", 
        "type": "Periodical"
      }, 
      {
        "issueNumber": "3", 
        "type": "PublicationIssue"
      }, 
      {
        "type": "PublicationVolume", 
        "volumeNumber": "55"
      }
    ], 
    "keywords": [
      "different frameworks", 
      "equalization effect", 
      "bagging", 
      "estimator", 
      "predictors", 
      "bootstrap samples", 
      "Bagged", 
      "reasons", 
      "success", 
      "occasional failure", 
      "failure", 
      "thanks", 
      "variance reduction", 
      "years", 
      "introduction", 
      "experimental evidence", 
      "evidence", 
      "hypothesis", 
      "prediction", 
      "influence", 
      "example", 
      "effect", 
      "framework", 
      "estimation", 
      "regression", 
      "improvement/deterioration", 
      "deterioration", 
      "situation", 
      "reduction argument", 
      "argument", 
      "strategies", 
      "identical effects", 
      "bootstrap sampling", 
      "samples", 
      "original predictors", 
      "effective thanks", 
      "reduction", 
      "training examples", 
      "real line", 
      "lines", 
      "goodness/badness", 
      "badness", 
      "influential examples", 
      "resampling strategy", 
      "sampling", 
      "paper", 
      "usual variance reduction argument", 
      "variance reduction argument"
    ], 
    "name": "Bagging Equalizes Influence", 
    "pagination": "251-270", 
    "productId": [
      {
        "name": "dimensions_id", 
        "type": "PropertyValue", 
        "value": [
          "pub.1013861997"
        ]
      }, 
      {
        "name": "doi", 
        "type": "PropertyValue", 
        "value": [
          "10.1023/b:mach.0000027783.34431.42"
        ]
      }
    ], 
    "sameAs": [
      "https://doi.org/10.1023/b:mach.0000027783.34431.42", 
      "https://app.dimensions.ai/details/publication/pub.1013861997"
    ], 
    "sdDataset": "articles", 
    "sdDatePublished": "2022-01-01T18:14", 
    "sdLicense": "https://scigraph.springernature.com/explorer/license/", 
    "sdPublisher": {
      "name": "Springer Nature - SN SciGraph project", 
      "type": "Organization"
    }, 
    "sdSource": "s3://com-springernature-scigraph/baseset/20220101/entities/gbq_results/article/article_395.jsonl", 
    "type": "ScholarlyArticle", 
    "url": "https://doi.org/10.1023/b:mach.0000027783.34431.42"
  }
]
 

Download the RDF metadata as:  json-ld nt turtle xml License info

HOW TO GET THIS DATA PROGRAMMATICALLY:

JSON-LD is a popular format for linked data which is fully compatible with JSON.

curl -H 'Accept: application/ld+json' 'https://scigraph.springernature.com/pub.10.1023/b:mach.0000027783.34431.42'

N-Triples is a line-based linked data format ideal for batch operations.

curl -H 'Accept: application/n-triples' 'https://scigraph.springernature.com/pub.10.1023/b:mach.0000027783.34431.42'

Turtle is a human-readable linked data format.

curl -H 'Accept: text/turtle' 'https://scigraph.springernature.com/pub.10.1023/b:mach.0000027783.34431.42'

RDF/XML is a standard XML format for linked data.

curl -H 'Accept: application/rdf+xml' 'https://scigraph.springernature.com/pub.10.1023/b:mach.0000027783.34431.42'


 

This table displays all metadata directly associated to this object as RDF triples.

118 TRIPLES      22 PREDICATES      77 URIs      66 LITERALS      6 BLANK NODES

Subject Predicate Object
1 sg:pub.10.1023/b:mach.0000027783.34431.42 schema:about anzsrc-for:17
2 anzsrc-for:1701
3 schema:author N36bdadef92b240b293d8276ac487a9bc
4 schema:citation sg:pub.10.1007/bf00058655
5 sg:pub.10.1023/a:1007515423169
6 sg:pub.10.1023/a:1007607513941
7 schema:datePublished 2004-06
8 schema:datePublishedReg 2004-06-01
9 schema:description Bagging constructs an estimator by averaging predictors trained on bootstrap samples. Bagged estimates almost consistently improve on the original predictor. It is thus important to understand the reasons for this success, and also for the occasional failures. It is widely believed that bagging is effective thanks to the variance reduction stemming from averaging predictors. However, seven years from its introduction, bagging is still not fully understood. This paper provides experimental evidence supporting the hypothesis that bagging stabilizes prediction by equalizing the influence of training examples. This effect is detailed in two different frameworks: estimation on the real line and regression. Bagging’s improvements/deteriorations are explained by the goodness/badness of highly influential examples, in situations where the usual variance reduction argument is at best questionable. Finally, reasons for the equalization effect are advanced. They support that other resampling strategies such as half-sampling should provide qualitatively identical effects while being computationally less demanding than bootstrap sampling.
10 schema:genre article
11 schema:inLanguage en
12 schema:isAccessibleForFree true
13 schema:isPartOf N463924465fff457da0aa8442ab841f36
14 Ne19f2cd65cf245978f2997675e94ce72
15 sg:journal.1125588
16 schema:keywords Bagged
17 argument
18 badness
19 bagging
20 bootstrap samples
21 bootstrap sampling
22 deterioration
23 different frameworks
24 effect
25 effective thanks
26 equalization effect
27 estimation
28 estimator
29 evidence
30 example
31 experimental evidence
32 failure
33 framework
34 goodness/badness
35 hypothesis
36 identical effects
37 improvement/deterioration
38 influence
39 influential examples
40 introduction
41 lines
42 occasional failure
43 original predictors
44 paper
45 prediction
46 predictors
47 real line
48 reasons
49 reduction
50 reduction argument
51 regression
52 resampling strategy
53 samples
54 sampling
55 situation
56 strategies
57 success
58 thanks
59 training examples
60 usual variance reduction argument
61 variance reduction
62 variance reduction argument
63 years
64 schema:name Bagging Equalizes Influence
65 schema:pagination 251-270
66 schema:productId Nc9935aeb50c2452c834a1c11b0c162ce
67 Ndc38d79f0f2844789733316076498c10
68 schema:sameAs https://app.dimensions.ai/details/publication/pub.1013861997
69 https://doi.org/10.1023/b:mach.0000027783.34431.42
70 schema:sdDatePublished 2022-01-01T18:14
71 schema:sdLicense https://scigraph.springernature.com/explorer/license/
72 schema:sdPublisher Nbc878c1f394f43ca8cb06f7863e501ef
73 schema:url https://doi.org/10.1023/b:mach.0000027783.34431.42
74 sgo:license sg:explorer/license/
75 sgo:sdDataset articles
76 rdf:type schema:ScholarlyArticle
77 N36bdadef92b240b293d8276ac487a9bc rdf:first sg:person.015255215731.52
78 rdf:rest rdf:nil
79 N463924465fff457da0aa8442ab841f36 schema:volumeNumber 55
80 rdf:type schema:PublicationVolume
81 Nbc878c1f394f43ca8cb06f7863e501ef schema:name Springer Nature - SN SciGraph project
82 rdf:type schema:Organization
83 Nc9935aeb50c2452c834a1c11b0c162ce schema:name doi
84 schema:value 10.1023/b:mach.0000027783.34431.42
85 rdf:type schema:PropertyValue
86 Ndc38d79f0f2844789733316076498c10 schema:name dimensions_id
87 schema:value pub.1013861997
88 rdf:type schema:PropertyValue
89 Ne19f2cd65cf245978f2997675e94ce72 schema:issueNumber 3
90 rdf:type schema:PublicationIssue
91 anzsrc-for:17 schema:inDefinedTermSet anzsrc-for:
92 schema:name Psychology and Cognitive Sciences
93 rdf:type schema:DefinedTerm
94 anzsrc-for:1701 schema:inDefinedTermSet anzsrc-for:
95 schema:name Psychology
96 rdf:type schema:DefinedTerm
97 sg:journal.1125588 schema:issn 0885-6125
98 1573-0565
99 schema:name Machine Learning
100 schema:publisher Springer Nature
101 rdf:type schema:Periodical
102 sg:person.015255215731.52 schema:affiliation grid-institutes:grid.462261.5
103 schema:familyName Grandvalet
104 schema:givenName Yves
105 schema:sameAs https://app.dimensions.ai/discover/publication?and_facet_researcher=ur.015255215731.52
106 rdf:type schema:Person
107 sg:pub.10.1007/bf00058655 schema:sameAs https://app.dimensions.ai/details/publication/pub.1002929950
108 https://doi.org/10.1007/bf00058655
109 rdf:type schema:CreativeWork
110 sg:pub.10.1023/a:1007515423169 schema:sameAs https://app.dimensions.ai/details/publication/pub.1017116781
111 https://doi.org/10.1023/a:1007515423169
112 rdf:type schema:CreativeWork
113 sg:pub.10.1023/a:1007607513941 schema:sameAs https://app.dimensions.ai/details/publication/pub.1041829946
114 https://doi.org/10.1023/a:1007607513941
115 rdf:type schema:CreativeWork
116 grid-institutes:grid.462261.5 schema:alternateName Heudiasyc, UMR CNRS 6599, Université de Technologie de Compiègne, France
117 schema:name Heudiasyc, UMR CNRS 6599, Université de Technologie de Compiègne, France
118 rdf:type schema:Organization
 




Preview window. Press ESC to close (or click here)


...