Methods, apparatus, and systems for estimating causal relationships between observed variables

Abstract

Methods, apparatus, and systems are disclosed for estimating causal relationships between observed variables. According to the method provided by the present disclosure, responding to the received observation data of the mixed observation variable, a mixed causal relationship target formula suitable for the continuous observation variable and the discrete observation variable is determined, whichincludes a causal relationship target formula for the continuous observation variable and a causal relationship target formula for the discrete observation variable, and the fitting inconsistency is adjusted based on a weighting factor of the observation variable. Then optimal solution is performed on the mixed causal relationship target expression by utilizing the mixed observation data and through mixed sparse causal reasoning suitable for continuous observation variables and discrete observation variables under the constraint of a directed acyclic graph so as to estimate the causal relationship among a plurality of observation variables. The embodiment of the invention is suitable for causal relationship estimation of the mixed observation variable, and the causal network structure haslow sensitivity to the observation variable estimation error, so that the accurate causal relationship can be obtained.

Description

technical field [0001] The present disclosure relates to the technical field of data mining, and more particularly to a method, device and system for estimating the causal relationship between observed variables. Background technique [0002] In the era of big data, a large amount of data can be obtained through various data collection channels. Through data analysis and mining of these data, a lot of useful information can be obtained from it. However, in many application fields, people often only see the appearance of the system, but cannot gain insight into the complex mechanism and process behind the system, but can only gain an empirical understanding. [0003] The causal structure learning is dedicated to the observation data based on the system, automatically restores the complex mechanism behind the system, and restores the data generation process. At present, causal structure learning technology has been applied in many fields such as pharmaceuticals, manufacturin...

AI Technical Summary

Problems solved by technology

[0008] However, the existing sparse causal modeling methods currently exist mainly for datasets where the data are both continuous variables or discrete variables, and for datasets where continuous variables and discrete variables coexist, existing causal models or inference algorithms either The time complexity is high, and it cannot support the learning of complex causal structure with high dimensionality of observed variables, or the results obtained due to the discretization of continuous variables and the judgment based on conditional independence are not optimal results, which in turn reduces the accuracy of causal structure learning

Images