A graphical design and implementation method for partially connecting a multi-layer perceptron

The invention discloses a graphical design and implementation method for partially connecting a multi-layer perceptron, comprising the following steps of: selecting a data set to be trained, and preprocessing data of the data set to meet the requirement of network training; Generating a corresponding data structure according to the network graph drawn by the user, and checking whether the connection mode of the network is legal or not; Performing topological sorting on the network diagram structure, executing forward calculation of the network according to the linear sequence, and calculatingan output value x on a neuron; According to the linear sequence obtained through topological sorting, executing reverse calculation of the network reversely, and calculating an error value on a neuron; Traversing directed edges in the network, and calculating weight values w of the gradient updated directed edges; The method can be used for training the partially-connected multilayer perceptron, the network is generated t