Counter-fact prediction method related to set type decision effect

Abstract

The invention provides a counter-fact prediction method about a set type decision effect, and belongs to the technical field of machine learning. According to the method, a decision variable and confusion variable decorrelation problem is converted into a lower-dimension decision variable implicit representation and confusion variable decorrelation problem, a probability density ratio estimation method based on a deep neural network, is adopted to take the probability density ratio of the joint distribution of the decision variable implicit representation and the confusion variable corresponding to the observation data sample to the joint distribution of the decision variable implicit representation and the confusion variable unassociated as the weight of a data point formed by the decision variable implicit representation and the confusion variable. A variational sample reweighting method is adopted to synthesize the weight of a data point formed by decision variable implicit representation and confusion variables into the weight of a sample in observation data, and the weighted observation data sample is utilized to train a counter-fact prediction model to perform counter-fact prediction on the effect of an individual under the influence of a specific decision. According to the invention, the accuracy of counter-fact prediction is improved, and the method has a very high application value.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and particularly proposes a counterfactual prediction method about the effect of set-type decision-making. Background technique [0002] Using a large amount of observational data, counterfactual prediction of the effects of heterogeneous individuals given different decisions is a problem of great significance in many fields. By predicting the effects of different decisions, it can help technicians in related fields to make more accurate decisions, such as selecting a series of processing procedures for a certain product in the field of industrial production, so that the quality of the product can reach the optimal situation. [0003] In order to predict the effects of different decisions, randomized controlled trials are the standard way to solve this kind of counterfactual prediction problem, that is, randomly assign decisions to research subjects and observe their effects. For exampl...

AI Technical Summary

Problems solved by technology

[0004] In order to de-confuse and bias the observed data, the existing counterfactual prediction technology for the effects of different decisions adopts the method of weighting the samples with the importance sampling weight, so that the decision variables in the weighted data are related to the confounding variables. decreased sex

The disadvantage of this type of technology is that it only solves the counterfactual prediction when the decision variable is a single variable, and the application range is too narrow

However, such an approach will cause problems of excessive complexity.

Images