Performance Comparison of Microsoft’s AutoML API

Philipp Neuhauser; Stefan Wagner

doi:10.46354/i3m.2023.emss.028

Performance Comparison of Microsoft’s AutoML API

Philipp Neuhauser, Stefan Wagner

Research output: Chapter in Book/Report/Conference proceedings › Conference contribution › peer-review

1 Citation (Scopus)

Abstract

In recent years, many software libraries for automated machine learning (AutoML), such as H2O AutoML or Auto-Sklearn, have become increasingly popular as they propose to significantly simplify the ML workflow and to sustainably reduce the time required for manual feature engineering, hyperparameter tuning as well as model selection and evaluation. Among the younger and therefore less well-known representatives is Microsoft’s ML.NET, about whose model builder API only little literature and performance comparisons exists. This paper summarizes the functionality of such frameworks and discusses general requirements for automated machine learning in more detail. Finally, several experiments compare ML.NET with already established AutoML libraries based on some datasets from the field of supervised learning with respect to model quality, the API’s scope of functions and required computational resources.

Translated title of the contribution	Performancevergleich von Microsofts AutoML API
Original language	English
Title of host publication	35th European Modeling and Simulation Symposium, EMSS 2023
Editors	Michael Affenzeller, Agostino G. Bruzzone, Emilio Jimenez, Francesco Longo, Antonella Petrillo
Publisher	DIME UNIVERSITY OF GENOA
ISBN (Electronic)	9788885741881
DOIs	https://doi.org/10.46354/i3m.2023.emss.028
Publication status	Published - 2023
Event	35th European Modeling and Simulation Symposium - Divani Caravel Hotel, Athen, Greece Duration: 18 Sept 2023 → 20 Sept 2023 https://www.msc-les.org/emss2023/

Publication series

Name	European Modeling and Simulation Symposium, EMSS
Volume	2023-September
ISSN (Print)	2305-2023

Conference

Conference	35th European Modeling and Simulation Symposium
Abbreviated title	EMSS 2023
Country/Territory	Greece
City	Athen
Period	18.09.2023 → 20.09.2023
Internet address	https://www.msc-les.org/emss2023/

Keywords

Automated Machine Learning
Microsoft AutoML
Performance Comparison
Hyperparameter Tuning

Access to Document

10.46354/i3m.2023.emss.028

Cite this

@inproceedings{e00ad0faf13f4a18942b20341a5ef33f,

title = "Performance Comparison of Microsoft{\textquoteright}s AutoML API",

abstract = "In recent years, many software libraries for automated machine learning (AutoML), such as H2O AutoML or Auto-Sklearn, have become increasingly popular as they propose to significantly simplify the ML workflow and to sustainably reduce the time required for manual feature engineering, hyperparameter tuning as well as model selection and evaluation. Among the younger and therefore less well-known representatives is Microsoft{\textquoteright}s ML.NET, about whose model builder API only little literature and performance comparisons exists. This paper summarizes the functionality of such frameworks and discusses general requirements for automated machine learning in more detail. Finally, several experiments compare ML.NET with already established AutoML libraries based on some datasets from the field of supervised learning with respect to model quality, the API{\textquoteright}s scope of functions and required computational resources.",

keywords = "Automated Machine Learning, Microsoft AutoML, Performance Comparison, Hyperparameter Tuning, Automated Machine Learning, Microsoft AutoML, Performance Comparison, Hyperparameter Tuning",

author = "Philipp Neuhauser and Stefan Wagner",

note = "Publisher Copyright: {\textcopyright} 2023 The Authors.; 35th European Modeling and Simulation Symposium ; Conference date: 18-09-2023 Through 20-09-2023",

year = "2023",

doi = "10.46354/i3m.2023.emss.028",

language = "English",

series = "European Modeling and Simulation Symposium, EMSS",

publisher = "DIME UNIVERSITY OF GENOA",

editor = "Michael Affenzeller and Bruzzone, {Agostino G.} and Emilio Jimenez and Francesco Longo and Antonella Petrillo",

booktitle = "35th European Modeling and Simulation Symposium, EMSS 2023",

url = "https://www.msc-les.org/emss2023/",

}

Neuhauser, P & Wagner, S 2023, Performance Comparison of Microsoft’s AutoML API. in M Affenzeller, AG Bruzzone, E Jimenez, F Longo & A Petrillo (eds), 35th European Modeling and Simulation Symposium, EMSS 2023. European Modeling and Simulation Symposium, EMSS, vol. 2023-September, DIME UNIVERSITY OF GENOA, 35th European Modeling and Simulation Symposium, Athen, Greece, 18.09.2023. https://doi.org/10.46354/i3m.2023.emss.028

Performance Comparison of Microsoft’s AutoML API. / Neuhauser, Philipp ; Wagner, Stefan.
35th European Modeling and Simulation Symposium, EMSS 2023. ed. / Michael Affenzeller; Agostino G. Bruzzone; Emilio Jimenez; Francesco Longo; Antonella Petrillo. DIME UNIVERSITY OF GENOA, 2023. (European Modeling and Simulation Symposium, EMSS; Vol. 2023-September).

Research output: Chapter in Book/Report/Conference proceedings › Conference contribution › peer-review

TY - GEN

T1 - Performance Comparison of Microsoft’s AutoML API

AU - Neuhauser, Philipp

AU - Wagner, Stefan

PY - 2023

Y1 - 2023

N2 - In recent years, many software libraries for automated machine learning (AutoML), such as H2O AutoML or Auto-Sklearn, have become increasingly popular as they propose to significantly simplify the ML workflow and to sustainably reduce the time required for manual feature engineering, hyperparameter tuning as well as model selection and evaluation. Among the younger and therefore less well-known representatives is Microsoft’s ML.NET, about whose model builder API only little literature and performance comparisons exists. This paper summarizes the functionality of such frameworks and discusses general requirements for automated machine learning in more detail. Finally, several experiments compare ML.NET with already established AutoML libraries based on some datasets from the field of supervised learning with respect to model quality, the API’s scope of functions and required computational resources.

AB - In recent years, many software libraries for automated machine learning (AutoML), such as H2O AutoML or Auto-Sklearn, have become increasingly popular as they propose to significantly simplify the ML workflow and to sustainably reduce the time required for manual feature engineering, hyperparameter tuning as well as model selection and evaluation. Among the younger and therefore less well-known representatives is Microsoft’s ML.NET, about whose model builder API only little literature and performance comparisons exists. This paper summarizes the functionality of such frameworks and discusses general requirements for automated machine learning in more detail. Finally, several experiments compare ML.NET with already established AutoML libraries based on some datasets from the field of supervised learning with respect to model quality, the API’s scope of functions and required computational resources.

KW - Automated Machine Learning

KW - Microsoft AutoML

KW - Performance Comparison

KW - Hyperparameter Tuning

KW - Automated Machine Learning

KW - Microsoft AutoML

KW - Performance Comparison

KW - Hyperparameter Tuning

UR - http://www.scopus.com/inward/record.url?scp=85179006926&partnerID=8YFLogxK

U2 - 10.46354/i3m.2023.emss.028

DO - 10.46354/i3m.2023.emss.028

M3 - Conference contribution

T3 - European Modeling and Simulation Symposium, EMSS

BT - 35th European Modeling and Simulation Symposium, EMSS 2023

A2 - Affenzeller, Michael

A2 - Bruzzone, Agostino G.

A2 - Jimenez, Emilio

A2 - Longo, Francesco

A2 - Petrillo, Antonella

PB - DIME UNIVERSITY OF GENOA

T2 - 35th European Modeling and Simulation Symposium

Y2 - 18 September 2023 through 20 September 2023

ER -

Performance Comparison of Microsoft’s AutoML API

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this