Online GBDT model learning method and device

Abstract

The invention provides an online GBDT prediction model learning method. The method comprises the steps of obtaining a sample data set used for learning a GBDT prediction model; and performing model learning based on the model parameters of at least one decision tree in the GBDT prediction model by using the sample data set to create a new decision tree and update the model parameters of the at least one decision tree, further, removing the partial decision tree from the at least one decision tree, and creating a new decision tree and updating the model parameters of the decision tree based onthe removed GBDT prediction model. By utilizing the method, the GBDT prediction model learning can be efficiently realized. In addition, a weight factor can be given to each decision tree of the GBDTprediction model, so that the model prediction accuracy is improved.

Description

technical field [0001] The present disclosure generally relates to the field of computer technology, and more specifically, relates to a method and device for online learning of a GBDT model. Background technique [0002] With the development of artificial intelligence, data mining and machine learning in the Internet and big data have become more and more important, and therefore many models for prediction have been produced to process the data to be predicted. Among various machine learning algorithms, the GBDT (Gradient Boosting Deision Tree, Gradient Boosting Decision Tree) algorithm is more and more widely used due to its excellent learning performance. The GBDT algorithm is a machine learning technique for regression, classification, sorting and other tasks, which obtains a strong prediction model by combining multiple weak learners (usually decision trees). [0003] figure 1 shows an example of a conventional GBDT prediction model consisting of T decision trees g 1...

AI Technical Summary

Problems solved by technology

[0003] figure 1 shows an example of a conventional GBDT prediction model consisting of T decision trees g 1 (x)~g T (x) constitutes, and the GBDT prediction model can be expressed as: Every time a new sample data set is used to train the GBDT prediction model, all decision trees g in the 1st to T-th decision trees need to be i (x) are all trained, which makes the amount of calculation large and consumes a lot of resources

Images