Abstract
Initializing the hyper-parameters (HPs) of machine learning (ML) techniques became an important step in the area of automated ML (AutoML). The main premise in HP initialization is that a HP setting that performs well for a certain dataset(s) will also be suitable for a similar dataset. Thus, evaluation of similarities of datasets based on their characteristics, named meta-features (MFs), is one of the basic tasks in meta-learning (MtL), a subfield of AutoML. Several types of MFs were developed from which those based on principal component analysis (PCA) are, despite their good descriptive characteristics and relatively easy computation, utilized only marginally. A novel approach to HP initialization combining dynamic time warping (DTW), a well-known similarity measure for time series, with PCA MFs is proposed in this paper which does not need any further settings. Exhaustive experiments, conducted for the use-cases of HP initialization of decision trees and support vector machines show the potential of the proposed approach and encourage further investigation in this direction.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
Since a meta-model is learned by traditional ML techniques.
- 2.
According to the vector \(\mathbf {m}^{c} = (\vartheta _1, \vartheta _2, \dots , \vartheta _c)\), \(d = min \{i\, | 1\le i \le c, \,\vartheta _i\ge 0.95\}\)), and c is the number of attributes in the dataset.
- 3.
Basically, a 10-bin histogram of the values of \(\mathbf {m}^{pca}\) from the Eq. 1.
- 4.
We are using all the eight MF types described above as well as their different combinations in our experiments as baselines.
- 5.
In case of MF vectors of the same size (e.g. \(\mathbf {m}^{his}\) and \(\mathbf {m}^{cup}\)) we were experimenting with well-known vector similarity measures such as Euclidean distance, inner product, cosine similarity and Pearson correlation.
- 6.
This study uses a simple average as an aggregation function, however, any other aggregation function, like a weighted average, can be used, too.
- 7.
- 8.
Where either all or none of the MFs belonging to a certain MF type were present in a MF vector, according to the given combination of MF types.
- 9.
This was done because of the stochastic nature of the used HP tuning algorithms (SMBO, PSO and RS), thus, to get more accurate statistics about the performance of the used approaches.
- 10.
References
Amasyali, M.F., Ersoy, O.K.: A study of meta learning for regression. Technical report, ECE Technical Reports. Paper 386, Purdue e-Pubs, Purdue University (2009)
Bardenet, R., Brendel, M., Kégl, B., Sebag, M.: Collaborative hyperparameter tuning. In: Proceedings of the 30th International Conference on Machine Learning, vol. 28, pp. 199–207. JMLR Workshop and Conference Proceedings (2013)
Bergstra, J., Bengio, Y.: Random search for hyper-parameter optimization. J. Mach. Learn. Res. 13, 281–305 (2012)
Berndt, D.J., Clifford, J.: Using dynamic time warping to find patterns in time series. In: AAAI Workshop on Knowledge Discovery in Databases, pp. 359–370 (1994)
Brazdil, P., Giraud-Carrier, C., Soares, C., Vilalta, R.: Metalearning: Applications to Data Mining, 1st edn. Springer, Heidelberg (2009)
Brodersen, K.H., Ong, C.S., Stephan, K.E., Buhmann, J.M.: The balanced accuracy and its posterior distribution. In: Proceedings of the 2010 20th International Conference on Pattern Recognition, pp. 3121–3124. IEEE Computer Society (2010)
Feurer, M., Klein, A., Eggensperger, K., Springenberg, J., Blum, M., Hutter, F.: Efficient and robust automated machine learning. In: Advances in Neural Information Processing Systems 28, pp. 2944–2952. Curran Associates, Inc. (2015)
Feurer, M., Springenberg, J.T., Hutter, F.: Using meta-learning to initialize bayesian optimization of hyperparameters. In: International Workshop on Meta-learning and Algorithm Selection co-located with 21st European Conference on Artificial Intelligence, pp. 3–10 (2014)
Janssens, J.H.: Outlier Selection and One-Class Classification. Ph.D. thesis, Tilburg University, Netherlands (2013), TiCC PhD Dissertation Series No.27
Mantovani, R.G., Rossi, A.L.D., Vanschoren, J., Bischl, B., de Carvalho, A.C.P.L.F.: To tune or not to tune: Recommending when to adjust svm hyper-parameters via meta-learning. In: 2015 International Joint Conference on Neural Networks, pp. 1–8 (2015)
Mantovani, R.G., Horváth, T., Cerri, R., Junior, S.B., de Vanschoren, J., de Carvalho, L.F.: An empirical study on hyperparameter tuning of decision trees, A.C.P. (2019)
Ratanamahatana, A., Keogh, E.: Everything you know about dynamic time warping is wrong. In: 3rd Workshop on Mining Temporal and Sequential Data, in conjunction with 10th ACM SIGKDD International Conference Knowledge Discovery and Data Mining (2004)
Sharma, A., Paliwal, K.K.: Fast principal component analysis using fixed-point algorithm. Pattern Recogn. Lett. 28(10), 1151–1155 (2007)
Snoek, J., Larochelle, H., Adams, R.P.: Practical bayesian optimization of machine learning algorithms. In: Pereira, F., Burges, C., Bottou, L., Weinberger, K. (eds.) Advances in Neural Information Processing Systems 25, pp. 2951–2959. Curran Associates, Inc. (2012)
Wistuba, M., Schilling, N., Schmidt-Thieme, L.: Hyperparameter search space pruning – a new component for sequential model-based hyperparameter optimization. In: Appice, A., Rodrigues, P.P., Santos Costa, V., Gama, J., Jorge, A., Soares, C. (eds.) ECML PKDD 2015. LNCS (LNAI), vol. 9285, pp. 104–119. Springer, Cham (2015). https://doi.org/10.1007/978-3-319-23525-7_7
Yang, X.S., Cui, Z., Xiao, R., Gandomi, A.H., Karamanoglu, M.: Swarm Intelligence and Bio-Inspired Computation: Theory and Applications, 1st edn. Elsevier Science Publishers B. V. (2013)
Acknowledgment
“Application Domain Specific Highly Reliable IT Solutions” project has been implemented with the support provided from the National Research, Development and Innovation Fund of Hungary, financed under the Thematic Excellence Programme TKP2020-NKA-06 (National Challenges Subprogramme) funding scheme.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this paper
Cite this paper
Horváth, T., Mantovani, R.G., de Carvalho, A.C.P.L.F. (2021). Time-Series in Hyper-parameter Initialization of Machine Learning Techniques. In: Yin, H., et al. Intelligent Data Engineering and Automated Learning – IDEAL 2021. IDEAL 2021. Lecture Notes in Computer Science(), vol 13113. Springer, Cham. https://doi.org/10.1007/978-3-030-91608-4_25
Download citation
DOI: https://doi.org/10.1007/978-3-030-91608-4_25
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-91607-7
Online ISBN: 978-3-030-91608-4
eBook Packages: Computer ScienceComputer Science (R0)