Auto-WEKA is a tool for automating the selection of methods and hyperparameters of WEKA. This repository contains an extended version of Auto-WEKA now supporting the optimisation of MultiComponent Predictive Systems (MCPS).
Many different machine learning algorithms exist that can easily be used off the shelf, many of these methods are implemented in the open source WEKA package. However, each of these algorithms have their own hyperparameters that can drastically change their performance, and there are a staggeringly large number of possible alternatives overall. Auto-WEKA considers the problem of simultaneously composing an MCPS and setting its hyperparameters, going beyond previous methods that address these issues in isolation. Auto-WEKA does this using a fully automated approach, leveraging recent innovations in Bayesian optimization. Our hope is that Auto-WEKA will help non-expert users to more effectively identify machine learning algorithms and hyperparameter settings appropriate to their applications, and hence to achieve improved performance.
Auto-WEKA includes a wizard to find the best MCPS of a given dataset without any user interaction, apart from providing a dataset and a time budget.
Auto-WEKA allows to select between a list of all WEKA methods and different optimisation strategies and parameters.
Clone the repository and run java -jar autoweka.jar. Then, follow userguide.pdf (please, note that PDF is from an older version. User guide for this new version will be updated soon).
- Manuel Martin Salvador, Marcin Budka, and Bogdan Gabrys. "Automatic composition and optimisation of multicomponent predictive systems with an extended Auto-WEKA" Submitted to IEEE Transactions on Automation Science and Engineering, 2017 (under review). [slides] [results]
- Manuel Martin Salvador, Marcin Budka, and Bogdan Gabrys. "Modelling Multi-Component Predictive Systems as Petri Nets". Submitted to 15th Annual Industrial Simulation Conference, 2017 (under review). [branch]
- Manuel Martin Salvador, Marcin Budka, and Bogdan Gabrys. "Effects of change propagation resulting from adaptive preprocessing in multicomponent predictive systems". In Proc. of the 20th International Conference KES-2016.
- Manuel Martin Salvador, Marcin Budka, and Bogdan Gabrys. "Adapting Multicomponent Predictive Systems using Hybrid Adaptation Strategies with Auto-WEKA in Process Industry". In Proc. of the 2016 Workshop on Automatic Machine Learning at ICML 2016. [branch] [results]
- Manuel Martin Salvador, Marcin Budka, and Bogdan Gabrys. "Towards automatic composition of multicomponent predictive systems" In Proc. of HAIS 2016, 2016. [slides] [results]
- Chris Thornton, Frank Hutter, Holger Hoos, and Kevin Leyton-Brown. "Auto-WEKA: Combined Selection and Hyperparameter Optimization of Classifiaction Algorithms" In Proc. of KDD 2013, 2013.
- Manuel Martin Salvador, PhD Candidate (Bournemouth University)
- Marcin Budka, Senior Lecturer (Bournemouth University)
- Bogdan Gabrys, Professor (Bournemouth University)
- Chris Thornton, M.Sc. Student (UBC)
- Frank Hutter, Assistant Professor (Freiburg University)
- Holger Hoos, Professor (UBC)
- Kevin Leyton-Brown, Associate Professor (UBC)
This version was developed using as base Auto-WEKA 0.5 from http://www.cs.ubc.ca/labs/beta/Projects/autoweka/
This software is intended for research purposes and not recommended for production environments. Support is not guaranteed, but please contact us if you have any question or would like to collaborate.
GNU General Public License v3 (see LICENSE)