.png){kind=icon}
.png){kind=icon}
Return
Compositional data graphical model learning based on conditional independence measurement
.png){kind=icon}
PRE
AI
DOI：10.1007/s11222-026-10821-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graphical models are important model tools in the fields of statistics and artificial intelligence. Developing graphical model learning methods for non-scalar data can enable their better application in various scenarios. Compositional data, as an important type of non-scalar data, are widely present in fields such as biology and social sciences. However, existing methods cannot address the problem of graphical model learning when all variables are compositional data of different dimensions. In this paper, we propose a graph-based measurement for conditional independence between compositional data variables, with values ranging between 0 and 1, capable of capturing both linear and nonlinear relationships. Based on this, we present a learning method for compositional data graphical models. A series of simulation experiments and real data analysis demonstrate the effectiveness of the proposed measurement and graphical model learning method, providing technical tools for the application of graphical models in the field of compositional data.
Keywords:
Compositional data
Graphical model
Conditional independence
Reproducing kernel Hilbert space
Remove constraints