OASC-2017: *Zilla Submission

Chris Cameron, Holger H. Hoos, Kevin Leyton-Brown, Frank Hutter
; Proceedings of the Open Algorithm Selection Challenge, PMLR 79:15-18, 2017.

Abstract

*Zilla is a model-based approach for algorithm selection and the most recent iteration of the well-known SATzilla project. The new *Zilla system has increased flexibility for the user and is configurable to run with many machine learning models and alternatives for various pre/post processing steps (e.g., presolver selection, feature completion prediction, and solver subset selection). The main additions to our *Zilla pipeline are automated procedures for feature group selection, hyper-parameter tuning, and solver subsampling prior to model building. We submit two versions for the competition that are equivalent except for the choice of per-instance machine learning model. For our first submission, we use a weighted pairwise random forest classifier. For our second submission, we test an experimental approach that offline, builds a weighted pairwise random forest classifier and online, finds the nearest instances based on the average path lengths across trees and optimizes a schedule over those instances.

Cite this Paper


BibTeX
@InProceedings{pmlr-v79-cameron17a, title = {OASC-2017: *Zilla Submission}, author = {Chris Cameron and Holger H. Hoos and Kevin Leyton-Brown and Frank Hutter}, booktitle = {Proceedings of the Open Algorithm Selection Challenge}, pages = {15--18}, year = {2017}, editor = {Marius Lindauer and Jan N. van Rijn and Lars Kotthoff}, volume = {79}, series = {Proceedings of Machine Learning Research}, address = {Brussels, Belgium}, month = {11--12 Sep}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v79/cameron17a/cameron17a.pdf}, url = {http://proceedings.mlr.press/v79/cameron17a.html}, abstract = {*Zilla is a model-based approach for algorithm selection and the most recent iteration of the well-known SATzilla project. The new *Zilla system has increased flexibility for the user and is configurable to run with many machine learning models and alternatives for various pre/post processing steps (e.g., presolver selection, feature completion prediction, and solver subset selection). The main additions to our *Zilla pipeline are automated procedures for feature group selection, hyper-parameter tuning, and solver subsampling prior to model building. We submit two versions for the competition that are equivalent except for the choice of per-instance machine learning model. For our first submission, we use a weighted pairwise random forest classifier. For our second submission, we test an experimental approach that offline, builds a weighted pairwise random forest classifier and online, finds the nearest instances based on the average path lengths across trees and optimizes a schedule over those instances.} }
Endnote
%0 Conference Paper %T OASC-2017: *Zilla Submission %A Chris Cameron %A Holger H. Hoos %A Kevin Leyton-Brown %A Frank Hutter %B Proceedings of the Open Algorithm Selection Challenge %C Proceedings of Machine Learning Research %D 2017 %E Marius Lindauer %E Jan N. van Rijn %E Lars Kotthoff %F pmlr-v79-cameron17a %I PMLR %J Proceedings of Machine Learning Research %P 15--18 %U http://proceedings.mlr.press %V 79 %W PMLR %X *Zilla is a model-based approach for algorithm selection and the most recent iteration of the well-known SATzilla project. The new *Zilla system has increased flexibility for the user and is configurable to run with many machine learning models and alternatives for various pre/post processing steps (e.g., presolver selection, feature completion prediction, and solver subset selection). The main additions to our *Zilla pipeline are automated procedures for feature group selection, hyper-parameter tuning, and solver subsampling prior to model building. We submit two versions for the competition that are equivalent except for the choice of per-instance machine learning model. For our first submission, we use a weighted pairwise random forest classifier. For our second submission, we test an experimental approach that offline, builds a weighted pairwise random forest classifier and online, finds the nearest instances based on the average path lengths across trees and optimizes a schedule over those instances.
APA
Cameron, C., Hoos, H.H., Leyton-Brown, K. & Hutter, F.. (2017). OASC-2017: *Zilla Submission. Proceedings of the Open Algorithm Selection Challenge, in PMLR 79:15-18

Related Material