• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
For visually-impairedUser profile (HSE staff only)Search

Open Source Library for Optimization and Regularization of Linear Predictive Classification Models on Large Data Sets

Student: Ilia Udalov

Supervisor: Konstantin V. Vorontsov

Faculty: Faculty of Computer Science

Educational Programme: System and Software Engineering (Master)

Final Grade: 9

Year of Graduation: 2017

This work is dedicated to development of open source library for optimization and regularization of linear predictive classification models on large data sets. Library uses naïve but fast algorithm to solve optimization problem unlike Support Vector Machine (SVM). Algorithm is based on iteration procedure and allows to use different optimizations, such as initial approximation (using Bayes classifier), multithreading implementation, introspective model adjustments while learning, choosing custom loss function. Algorithm also saves SVM’like generalization for nonlinear cases with kernel trick. Theoretical part of this work studies structure, complexity and behavior of this approach for learning linear predictive classification models. Practical outcome of work is an open source library. Library has C++ API and command line interface. Library is publicly available on GitHub. This library was covered by tests and compared with most popular SVM implementation on multiple public data sets. ROC-AUC metric is used to compare models. Test datasets represent different tasks with different complexity from classic Fischer’s Iris data set to classifying internet advert, cancer, materials, etc. Keywords: linear predictive classification models, regularization, optimization methods, kernel trick, minimum empirical risk principle, support vector machine Master thesis: 49 pages, 27 pictures, 2 tables, 22 sources

Full text (added June 5, 2017)

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses