Abstract: End-to-end gradient boosting network is a neural network composed of multiple base learners. It is similar to the structure of residual network. The behind network units （learners or residual blocks） in the two networks continue to learn on the basis of the former network units to gradually approach the objective function. The network structure of end-to-end gradient boosting network is more complex， and we have not enough understanding of its working mechanism. Visualization technology is helpful for us to intuitively understand the working mechanism within the network. This paper focuses on exploring the classification process and characteristics of end-to-end gradient boosting network. We visualized the classification process of it on simulated data， highlights its characteristics and problems by comparing with fully connected network and residual network. We use dummy nodes to show that its self regularization ability is relatively weak. Then， the visual method is used to explore the influence of learning rate on the classification process. Finally， through the experiment on the actual classification task， the correctness of the visualization related conclusions is verified to a certain extent.