Stability selection for component-wise gradient boosting in multiple dimensions
We present a new algorithm for boosting generalized additive models for location, scale and shape (GAMLSS) that allows to incorporate stability selection, an increasingly popular way to obtain stable sets of covariates while controlling the per-family error rate (PFER). The model is fitted repeatedl...
Main Authors: | , , , , , |
---|---|
Format: | Text |
Language: | unknown |
Published: |
arXiv
2016
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.1611.10171 https://arxiv.org/abs/1611.10171 |
id |
ftdatacite:10.48550/arxiv.1611.10171 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.1611.10171 2023-05-15T15:55:57+02:00 Stability selection for component-wise gradient boosting in multiple dimensions Thomas, Janek Mayr, Andreas Bischl, Bernd Schmid, Matthias Smith, Adam Hofner, Benjamin 2016 https://dx.doi.org/10.48550/arxiv.1611.10171 https://arxiv.org/abs/1611.10171 unknown arXiv https://dx.doi.org/10.1007/s11222-017-9754-6 arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ Computation stat.CO Machine Learning stat.ML FOS Computer and information sciences article-journal Article ScholarlyArticle Text 2016 ftdatacite https://doi.org/10.48550/arxiv.1611.10171 https://doi.org/10.1007/s11222-017-9754-6 2022-04-01T11:18:20Z We present a new algorithm for boosting generalized additive models for location, scale and shape (GAMLSS) that allows to incorporate stability selection, an increasingly popular way to obtain stable sets of covariates while controlling the per-family error rate (PFER). The model is fitted repeatedly to subsampled data and variables with high selection frequencies are extracted. To apply stability selection to boosted GAMLSS, we develop a new "noncyclical" fitting algorithm that incorporates an additional selection step of the best-fitting distribution parameter in each iteration. This new algorithms has the additional advantage that optimizing the tuning parameters of boosting is reduced from a multi-dimensional to a one-dimensional problem with vastly decreased complexity. The performance of the novel algorithm is evaluated in an extensive simulation study. We apply this new algorithm to a study to estimate abundance of common eider in Massachusetts, USA, featuring excess zeros, overdispersion, non-linearity and spatio-temporal structures. Eider abundance is estimated via boosted GAMLSS, allowing both mean and overdispersion to be regressed on covariates. Stability selection is used to obtain a sparse set of stable predictors. : 16 pages Text Common Eider DataCite Metadata Store (German National Library of Science and Technology) |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
unknown |
topic |
Computation stat.CO Machine Learning stat.ML FOS Computer and information sciences |
spellingShingle |
Computation stat.CO Machine Learning stat.ML FOS Computer and information sciences Thomas, Janek Mayr, Andreas Bischl, Bernd Schmid, Matthias Smith, Adam Hofner, Benjamin Stability selection for component-wise gradient boosting in multiple dimensions |
topic_facet |
Computation stat.CO Machine Learning stat.ML FOS Computer and information sciences |
description |
We present a new algorithm for boosting generalized additive models for location, scale and shape (GAMLSS) that allows to incorporate stability selection, an increasingly popular way to obtain stable sets of covariates while controlling the per-family error rate (PFER). The model is fitted repeatedly to subsampled data and variables with high selection frequencies are extracted. To apply stability selection to boosted GAMLSS, we develop a new "noncyclical" fitting algorithm that incorporates an additional selection step of the best-fitting distribution parameter in each iteration. This new algorithms has the additional advantage that optimizing the tuning parameters of boosting is reduced from a multi-dimensional to a one-dimensional problem with vastly decreased complexity. The performance of the novel algorithm is evaluated in an extensive simulation study. We apply this new algorithm to a study to estimate abundance of common eider in Massachusetts, USA, featuring excess zeros, overdispersion, non-linearity and spatio-temporal structures. Eider abundance is estimated via boosted GAMLSS, allowing both mean and overdispersion to be regressed on covariates. Stability selection is used to obtain a sparse set of stable predictors. : 16 pages |
format |
Text |
author |
Thomas, Janek Mayr, Andreas Bischl, Bernd Schmid, Matthias Smith, Adam Hofner, Benjamin |
author_facet |
Thomas, Janek Mayr, Andreas Bischl, Bernd Schmid, Matthias Smith, Adam Hofner, Benjamin |
author_sort |
Thomas, Janek |
title |
Stability selection for component-wise gradient boosting in multiple dimensions |
title_short |
Stability selection for component-wise gradient boosting in multiple dimensions |
title_full |
Stability selection for component-wise gradient boosting in multiple dimensions |
title_fullStr |
Stability selection for component-wise gradient boosting in multiple dimensions |
title_full_unstemmed |
Stability selection for component-wise gradient boosting in multiple dimensions |
title_sort |
stability selection for component-wise gradient boosting in multiple dimensions |
publisher |
arXiv |
publishDate |
2016 |
url |
https://dx.doi.org/10.48550/arxiv.1611.10171 https://arxiv.org/abs/1611.10171 |
genre |
Common Eider |
genre_facet |
Common Eider |
op_relation |
https://dx.doi.org/10.1007/s11222-017-9754-6 |
op_rights |
arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
op_doi |
https://doi.org/10.48550/arxiv.1611.10171 https://doi.org/10.1007/s11222-017-9754-6 |
_version_ |
1766391423041536000 |