Adaptive exogenous variable identification method

Abstract

The invention discloses an adaptive exogenous variable identification method, comprising the following steps: setting a data set, wherein each variable in the data set contains m sample data, setting a matrix, setting an array; calculating the data set and other Perform least squares regression operation on all data to obtain residuals; calculate the mutual information between variables and all residuals; replace the mutual information with the elements in the matrix; calculate the maximum value of each row in the matrix and store it in the array; Find the minimum value in the array; find the variable that is independent of all remaining residuals is the exogenous variable. The invention utilizes the idea of ​​maximum and minimum, combined with the characteristics of exogenous variables, so that the introduced independence determination parameter is an adaptive parameter value, which avoids the problem of low recognition rate caused by traditional algorithms being sensitive to differences in independence values, and also avoids the Different data sets are sensitive to fixed independence parameters and lead to unidentifiable defects, improving the identification of exogenous variables.

Description

technical field [0001] The invention relates to the technical field of data mining, and more specifically relates to an adaptive exogenous variable identification method. Background technique [0002] The causal discovery algorithm is mainly widely used in the field of artificial intelligence. The so-called causal discovery algorithm is an algorithm based on the identification of exogenous variables to generate a reaction action mechanism. From the definition of the causal discovery algorithm above, the exogenous variable is The trigger of the causal discovery algorithm, correctly identifying exogenous variables provides effective intervention measures in the process of artificial intelligence control, so that the artificial intelligence system can better understand the causal mechanism between things. [0003] There are two main methods for identifying exogenous variables in the prior art, both of which are based on the principle of linear non-Gaussian. The first identifica...

AI Technical Summary

Problems solved by technology

[0004] The disadvantage of the first method is that it will need to add certain assumptions and restrictions, and use non-Gaussian indicators to identify exogenous variables. Therefore, the first identification method can only accurately identify certain types of data and is not universal; the second method The disadvantage is that taking the variable corresponding to the minimum value as an exogenous variable in a summation method is an unreliable measure

[0005] In summary, using the above two methods to identify exogenous variables has the disadvantages of insufficient adaptability and reliability

Images