Javascript must be enabled for the correct page display

Cross-comparison of inference methods for Gaussian Graphical Models

Schotanus, Sijbren (2018) Cross-comparison of inference methods for Gaussian Graphical Models. Bachelor's Thesis, Mathematics.

[bMATH]_[2018]_[Sijbren Schotanus].pdf.pdf

Download (831kB) | Preview
[img] Text
Restricted to Registered users only

Download (94kB)


In Gaussian graphical modeling statistics and graph theory go hand in hand. With applications in many fields this widely applied technique is used to estimate the partial correlations among variables of a given dataset. This thesis covers four of the popular approaches within Gaussian graphical modeling, the Moore-Penrose pseudoinverse, the Lasso (inclusion of the `1-term), the shrinkage and the bootstrap approach. These four methods will be cross compared on the basis of several tests. These tests cover the accuracy of the estimation, the computational speed and the user friendliness of each method. The first test is based on randomly generated datasets of which the true partial correlations are know. Every tested method is then applied to two datasets extracted from the R package ’GeneNet’. In the last section of this thesis it is concluded that, despite a high demand on the computational cost, the Lasso approach supplies the user with the most accurate estimation. However it has to be noted that it is suspected that the shrinkage approach did not achieve its full potential due to mistakes made along the journey of programming.

Item Type: Thesis (Bachelor's Thesis)
Supervisor name: Grzegorczyk, M.A.
Degree programme: Mathematics
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 23 Aug 2018
Last Modified: 04 Sep 2018 13:53

Actions (login required)

View Item View Item