|LJK - Bâtiment IMAG|
|700 avenue Centrale|
|Campus St Martin d'Hères|
|38401 Grenoble France|
|Supervisors:||Kévin Polisano, Sophie Achard, Irène Gannaz|
Graphical models generation
Graph inference is an area which encounters much attention. The objective is to represent the dependence between individuals or variables as a graph. The first advantage is visual.
It is also attractive because it gives access to many tools from graph theory, e.g. for structure comparisons. A graphical model is a modeling of the dependence structure based on a covariance matrix or its inverse (precision matrix). Validation of inference procedures needs a simulation study. Yet, the generation of covariance matrices encounters obstacles.
In particular, the values of generated correlation matrices are often low, compared with real life observations. Recently, Cordoba et al. (2020) proposed a new procedure for generating covariance matrices. Our objective is to do an overview of existing methods in this domain, starting from this article, and then evaluate numerically the quality of the methods.