US 10,706,104 B1Granted

System and method for generating a graphical model

AssigneeBabylon Partners Limited

InventorsCiaran M. Lee, Anish Dhir

Life of this patent

Estimating the term…

Loading patent…

Abstract

Methods for creating a model that is a graphical representation are provided. In one aspect, a method includes including receiving a first dataset including a first variable and a third variable, and a second dataset including a second variable and the third variable. The method also includes creating graphical representations of the first and second datasets by applying conditional independence tests on them, and storing conditional independence information obtained by applying the conditional independence tests on the first and second datasets. The method also includes applying a bivariate causal discovery algorithm. The method further includes modifying the graphical representations of the first and second dataset according to the determined causal relations, and creating a set of candidate graphical representations for a third dataset including the first and second datasets. Each candidate graphical representation is consistent with the conditional independence information. Systems and machine-readable media are also provided.

Claims13

Dependent claims

A computer implemented method of creating a model that is a graphical representation comprising:

receiving a first dataset comprising a first variable and a third variable and a second dataset comprising a second variable and the third variable, wherein the first dataset and second dataset are overlapping in the third variable, wherein the first dataset and second dataset were not measured at the same time;
creating graphical representations of the first dataset and the second dataset by applying conditional independence tests on the first dataset and second dataset;
storing conditional independence information obtained by applying the conditional independence tests on the first dataset and the second dataset;
applying a bivariate causal discovery algorithm to determine a causal relation between the first and third variable in the first dataset and a causal relation between the second and third variable in the second dataset, the causal discovery algorithm being able to determine if the first variable causes the third variable, the third variable causes the first variable, the second variable causes the third variable and the third variable causes the second variable;
modifying the graphical representations of the first and second dataset according to the determined causal relations;
and
creating a set of candidate graphical representations for a third dataset comprising the first dataset and the second dataset, wherein each candidate graphical representation indicates a causal relationship between the first variable and the second variable, wherein each candidate graphical representation is consistent with the conditional independence information.

4 more elements

A system adapted to generate a model that is a graphical representation, the system comprising a processor and a memory, the processor being adapted to:

receive a first dataset comprising a first variable and a third variable and a second dataset comprising a second variable and the third variable, wherein the first dataset and second dataset are overlapping in the third variable, wherein the first dataset and second dataset were not measured at the same time;
create graphical representations of the first dataset and the second dataset by applying conditional independence tests on the first dataset and second dataset;
store conditional independence information obtained by applying the conditional independence tests on the first dataset and the second dataset;
apply a bivariate causal discovery algorithm to determine a causal relation between the first and third variable in the first dataset and a causal relation between the second and third variable in the second dataset, the causal discovery algorithm being able to determine if the first variable causes the third variable, the third variable causes the first variable, the second variable causes the third variable and the third variable causes the second variable;
modify the graphical representations of the first and second dataset according to the determined causal relations;
and
create a set of candidate graphical representations for a third dataset comprising the first dataset and the second dataset, wherein each candidate graphical representation indicates a causal relationship between the first variable and the second variable, wherein each candidate graphical representation is consistent with the conditional independence information.

4 more elements

Dependencies are inferred from the claim text.