Longitudinal training method and device of GBDT model, electronic equipment and system

Abstract

The invention discloses a longitudinal training method and device for a GBDT model, electronic equipment and a system, and the method comprises the steps: obtaining node numbers of nodes, located in a current layer, of a plurality of samples for the current layer of a current decision tree of an initial GBDT model; for each node of the current layer, binning all samples included in the node according to the binning number of each feature held by the node based on the node numbers of the plurality of samples to obtain a binning result; aiming at various splitting standards which can be used for performing node splitting of the current layer, respectively calculating a score gain value for performing node splitting according to each splitting standard, and taking the splitting standard with the highest score gain value as the splitting standard of the current layer; and according to the splitting standard, performing node splitting to obtain a node splitting result of the current layer, and completing GBDT model training. By adopting the scheme, the longitudinal training for the GBDT model is realized under the condition that each data node does not leak own feature data.

Description

technical field [0001] The present application relates to the field of machine learning technology and the field of multi-party secure computing technology, in particular to a longitudinal training method, device, electronic equipment and system of a GBDT model. Background technique [0002] GBDT (Gradient Boosting Decision Tree, Gradient Boosting Decision Tree) is a set of supervised learning algorithms that use decision tree technology for model training. By fitting multiple decision trees, the prediction result of the model is close to the real value used during training. The target value fitted by each decision tree is equal to the difference between the real value of the training set and the predicted value of the previous decision trees. . The GBDT model is usually applied to problems such as classification and regression. [0003] At present, in the practical application of the GBDT model, the feature data of various features of multiple samples when training the GB...

AI Technical Summary

Problems solved by technology

[0003] At present, in the practical application of the GBDT model, the feature data of various features of multiple samples when training the GBDT model are held by different participants, and these participants do not want to leak multiple samples The feature data of the features held by itself, resulting in the inability to achieve effective training of the GBDT model

Images